From 6d56d1415e24df3693d071352098630ebf18adf6 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 16:16:40 -0700 Subject: Update button style in mnt notebook --- .../python/examples/nmt_with_attention/nmt_with_attention.ipynb | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb b/tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb index 54ebcad8e9..34ce5e0cc3 100644 --- a/tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb +++ b/tensorflow/contrib/eager/python/examples/nmt_with_attention/nmt_with_attention.ipynb @@ -41,11 +41,11 @@ "\n", "# Neural Machine Translation with Attention\n", "\n", - "
\n", + "
\n", "\n", - " Run in Google Colab \n", + " Run in Google Colab \n", "\n", - "View source on Github
" + "View source on GitHub
" ] }, { -- cgit v1.2.3 From e0e07a338708a7052713c6244d7a44073b624e4b Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 16:30:59 -0700 Subject: Update eager notebooks with colab/github buttons. Change titles. Open leftnav. --- .../eager/python/examples/notebooks/1_basics.ipynb | 350 ++++++++++------- .../python/examples/notebooks/2_gradients.ipynb | 252 ++++++++----- .../python/examples/notebooks/3_datasets.ipynb | 208 +++++++---- .../examples/notebooks/3_training_models.ipynb | 380 +++++++++++-------- .../python/examples/notebooks/4_high_level.ipynb | 416 ++++++++++++--------- 5 files changed, 953 insertions(+), 653 deletions(-) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb index 51d10a7784..99d141bf2b 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb @@ -1,27 +1,101 @@ { + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Eager execution basics", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, "cells": [ { + "metadata": { + "id": "iPpI7RaYoZuE", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { "metadata": { - "colab_type": "text", - "id": "U9i2Dsh-ziXr" + "id": "hro2InpHobKk", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Eager execution basics" + ] + }, + { + "metadata": { + "id": "Hndw-YcxoOJK", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "6sILUVbHoSgH", + "colab_type": "text" + }, + "cell_type": "markdown", "source": [ - "# An introduction to TensorFlow\n", - "\n", "This is an introductory tutorial for using TensorFlow. It will cover:\n", "\n", "* Importing required packages\n", "* Creating and using Tensors\n", - "* Using GPU acceleration\n" + "* Using GPU acceleration" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "z1JcS5iBXMRO" + "id": "z1JcS5iBXMRO", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Import TensorFlow\n", "\n", @@ -30,32 +104,32 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { - "cellView": "code", + "id": "RlIWhyeLoYnG", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } }, - "colab_type": "code", - "id": "RlIWhyeLoYnG" + "cellView": "code" }, - "outputs": [], + "cell_type": "code", "source": [ "import tensorflow as tf\n", "\n", "tf.enable_eager_execution()" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "H9UySOPLXdaw" + "id": "H9UySOPLXdaw", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Tensors\n", "\n", @@ -63,10 +137,9 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { - "cellView": "code", + "id": "ngUe237Wt48W", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -74,7 +147,8 @@ }, "height": 125 }, - "colab_type": "code", + "cellView": "code", + "outputId": "b1a1cd60-4eb3-443d-cd6b-68406390784e", "executionInfo": { "elapsed": 320, "status": "ok", @@ -85,13 +159,22 @@ "userId": "" }, "user_tz": 420 - }, - "id": "ngUe237Wt48W", - "outputId": "b1a1cd60-4eb3-443d-cd6b-68406390784e" + } }, + "cell_type": "code", + "source": [ + "print(tf.add(1, 2))\n", + "print(tf.add([1, 2], [3, 4]))\n", + "print(tf.square(5))\n", + "print(tf.reduce_sum([1, 2, 3]))\n", + "print(tf.encode_base64(\"hello world\"))\n", + "\n", + "# Operator overloading is also supported\n", + "print(tf.square(2) + tf.square(3))" + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "tf.Tensor(3, shape=(), dtype=int32)\n", @@ -100,34 +183,25 @@ "tf.Tensor(6, shape=(), dtype=int32)\n", "tf.Tensor(aGVsbG8gd29ybGQ, shape=(), dtype=string)\n", "tf.Tensor(13, shape=(), dtype=int32)\n" - ] + ], + "name": "stdout" } - ], - "source": [ - "print(tf.add(1, 2))\n", - "print(tf.add([1, 2], [3, 4]))\n", - "print(tf.square(5))\n", - "print(tf.reduce_sum([1, 2, 3]))\n", - "print(tf.encode_base64(\"hello world\"))\n", - "\n", - "# Operator overloading is also supported\n", - "print(tf.square(2) + tf.square(3))" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "IDY4WsYRhP81" + "id": "IDY4WsYRhP81", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Each Tensor has a shape and a datatype" ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "srYWH1MdJNG7", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -135,7 +209,7 @@ }, "height": 53 }, - "colab_type": "code", + "outputId": "5e4ac41c-5115-4e50-eba0-42e249c16561", "executionInfo": { "elapsed": 215, "status": "ok", @@ -146,32 +220,32 @@ "userId": "" }, "user_tz": 420 - }, - "id": "srYWH1MdJNG7", - "outputId": "5e4ac41c-5115-4e50-eba0-42e249c16561" + } }, + "cell_type": "code", + "source": [ + "x = tf.matmul([[1]], [[2, 3]])\n", + "print(x.shape)\n", + "print(x.dtype)" + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "(1, 2)\n", - "\u003cdtype: 'int32'\u003e\n" - ] + "\n" + ], + "name": "stdout" } - ], - "source": [ - "x = tf.matmul([[1]], [[2, 3]])\n", - "print(x.shape)\n", - "print(x.dtype)" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "eBPw8e8vrsom" + "id": "eBPw8e8vrsom", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", "\n", @@ -180,11 +254,11 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "Dwi1tdW3JBw6" + "id": "Dwi1tdW3JBw6", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### NumPy Compatibility\n", "\n", @@ -197,9 +271,9 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -207,7 +281,7 @@ }, "height": 251 }, - "colab_type": "code", + "outputId": "fd0a22bc-8249-49dd-fcbd-63161cc47e46", "executionInfo": { "elapsed": 238, "status": "ok", @@ -218,13 +292,28 @@ "userId": "" }, "user_tz": 420 - }, - "id": "lCUWzso6mbqR", - "outputId": "fd0a22bc-8249-49dd-fcbd-63161cc47e46" + } }, + "cell_type": "code", + "source": [ + "import numpy as np\n", + "\n", + "ndarray = np.ones([3, 3])\n", + "\n", + "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", + "tensor = tf.multiply(ndarray, 42)\n", + "print(tensor)\n", + "\n", + "\n", + "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", + "print(np.add(tensor, 1))\n", + "\n", + "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", + "print(tensor.numpy())" + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "TensorFlow operations convert numpy arrays to Tensors automatically\n", @@ -240,32 +329,17 @@ "[[ 42. 42. 42.]\n", " [ 42. 42. 42.]\n", " [ 42. 42. 42.]]\n" - ] + ], + "name": "stdout" } - ], - "source": [ - "import numpy as np\n", - "\n", - "ndarray = np.ones([3, 3])\n", - "\n", - "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", - "tensor = tf.multiply(ndarray, 42)\n", - "print(tensor)\n", - "\n", - "\n", - "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", - "print(np.add(tensor, 1))\n", - "\n", - "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", - "print(tensor.numpy())" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "PBNP8yTRfu_X" + "id": "PBNP8yTRfu_X", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## GPU acceleration\n", "\n", @@ -273,10 +347,9 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { - "cellView": "code", + "id": "3Twf_Rw-gQFM", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -284,7 +357,8 @@ }, "height": 53 }, - "colab_type": "code", + "cellView": "code", + "outputId": "2239ae2b-adf3-4895-b1f3-464cf5361d1b", "executionInfo": { "elapsed": 340, "status": "ok", @@ -295,20 +369,9 @@ "userId": "" }, "user_tz": 420 - }, - "id": "3Twf_Rw-gQFM", - "outputId": "2239ae2b-adf3-4895-b1f3-464cf5361d1b" - }, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Is there a GPU available: False\n", - "Is the Tensor on GPU #0: False\n" - ] } - ], + }, + "cell_type": "code", "source": [ "x = tf.random_uniform([3, 3])\n", "\n", @@ -317,26 +380,37 @@ "\n", "print(\"Is the Tensor on GPU #0: \"),\n", "print(x.device.endswith('GPU:0'))" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "Is there a GPU available: False\n", + "Is the Tensor on GPU #0: False\n" + ], + "name": "stdout" + } ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "vpgYzgVXW2Ud" + "id": "vpgYzgVXW2Ud", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Device Names\n", "\n", - "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:\u003cN\u003e` if the tensor is placed on the `N`-th tensor on the host." + "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "ZWZQCimzuqyP" + "id": "ZWZQCimzuqyP", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "\n", "\n", @@ -346,9 +420,9 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "RjkNZTuauy-Q", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -356,7 +430,7 @@ }, "height": 53 }, - "colab_type": "code", + "outputId": "2e613293-ccac-4db2-b793-8ceb5b5adcfd", "executionInfo": { "elapsed": 1762, "status": "ok", @@ -367,20 +441,9 @@ "userId": "" }, "user_tz": 420 - }, - "id": "RjkNZTuauy-Q", - "outputId": "2e613293-ccac-4db2-b793-8ceb5b5adcfd" - }, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "On CPU:\n", - "10 loops, best of 3: 35.8 ms per loop\n" - ] } - ], + }, + "cell_type": "code", "source": [ "def time_matmul(x):\n", " %timeit tf.matmul(x, x)\n", @@ -398,14 +461,25 @@ " x = tf.random_uniform([1000, 1000])\n", " assert x.device.endswith(\"GPU:0\")\n", " time_matmul(x)" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "On CPU:\n", + "10 loops, best of 3: 35.8 ms per loop\n" + ], + "name": "stdout" + } ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "YEOJTNiOvnpQ" + "id": "YEOJTNiOvnpQ", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Next Steps\n", "\n", @@ -413,17 +487,5 @@ "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." ] } - ], - "metadata": { - "colab": { - "collapsed_sections": [], - "default_view": {}, - "name": "TensorFlow: An introduction", - "provenance": [], - "version": "0.3.2", - "views": {} - } - }, - "nbformat": 4, - "nbformat_minor": 0 -} + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb index 9c1af9c208..d7bcf749e8 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb @@ -1,54 +1,128 @@ { + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Automatic differentiation and gradient tape", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, "cells": [ { + "metadata": { + "id": "t09eeeR5prIJ", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { "metadata": { - "colab_type": "text", - "id": "vDJ4XzMqodTy" + "id": "GCCk8_dHpuNf", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" }, + "cell_type": "code", "source": [ - "# Automatic Differentiation\n", - "\n", - "In the previous tutorial we introduced `Tensor`s and operations on them. In this tutorial we will cover [automatic differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation), a key technique for optimizing machine learning models." + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "xh8WkEwWpnm7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Automatic differentiation and gradient tape" ] }, { + "metadata": { + "id": "idv0bPeCp325", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { "metadata": { - "colab_type": "text", - "id": "GQJysDM__Qb0" + "id": "vDJ4XzMqodTy", + "colab_type": "text" }, + "cell_type": "markdown", + "source": [ + "In the previous tutorial we introduced `Tensor`s and operations on them. In this tutorial we will cover [automatic differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation), a key technique for optimizing machine learning models." + ] + }, + { + "metadata": { + "id": "GQJysDM__Qb0", + "colab_type": "text" + }, + "cell_type": "markdown", "source": [ "## Setup\n" ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "OiMPZStlibBv", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "OiMPZStlibBv" + } }, - "outputs": [], + "cell_type": "code", "source": [ "import tensorflow as tf\n", "tf.enable_eager_execution()\n", "\n", "tfe = tf.contrib.eager # Shorthand for some symbols" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "1CLWJl0QliB0" + "id": "1CLWJl0QliB0", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Derivatives of a function\n", "\n", @@ -56,19 +130,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "9FViq92UX7P8", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "9FViq92UX7P8" + } }, - "outputs": [], + "cell_type": "code", "source": [ "from math import pi\n", "\n", @@ -82,15 +154,17 @@ "# with respect to its arguments. Since f() has a single argument,\n", "# grad_f will return a list with a single element.\n", "grad_f = tfe.gradients_function(f)\n", - "assert tf.abs(grad_f(pi/2)[0]).numpy() \u003c 1e-7" - ] + "assert tf.abs(grad_f(pi/2)[0]).numpy() < 1e-7" + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "v9fPs8RyopCf" + "id": "v9fPs8RyopCf", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Higher-order gradients\n", "\n", @@ -98,9 +172,9 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "3D0ZvnGYo0rW", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -108,7 +182,7 @@ }, "height": 276 }, - "colab_type": "code", + "outputId": "e23f8cc6-6813-4944-f20f-825b8a03c2ff", "executionInfo": { "elapsed": 730, "status": "ok", @@ -119,24 +193,9 @@ "userId": "" }, "user_tz": 420 - }, - "id": "3D0ZvnGYo0rW", - "outputId": "e23f8cc6-6813-4944-f20f-825b8a03c2ff" - }, - "outputs": [ - { - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEDCAYAAAAhsS8XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nF3bGGIswXNgZYyzC/Be68EGj7hfMcwAAAABJRU5ErkJggg==\n", - "text/plain": [ - "\u003cmatplotlib.figure.Figure at 0x7f385e198650\u003e" - ] - }, - "metadata": { - "tags": [] - }, - "output_type": "display_data" } - ], + }, + "cell_type": "code", "source": [ "def f(x):\n", " return tf.square(tf.sin(x))\n", @@ -154,14 +213,29 @@ "plt.plot(x, grad(grad(grad(f)))(x), label=\"third derivative\")\n", "plt.legend()\n", "plt.show()" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "display_data", + "data": { + "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEDCAYAAAAhsS8XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nF3bGGIswXNgZYyzC/Be68EGj7hfMcwAAAABJRU5ErkJggg==\n", + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + } + } ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "-39gouo7mtgu" + "id": "-39gouo7mtgu", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Gradient tapes\n", "\n", @@ -172,19 +246,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "MH0UfjympWf7", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "MH0UfjympWf7" + } }, - "outputs": [], + "cell_type": "code", "source": [ "def f(x, y):\n", " output = 1\n", @@ -200,14 +272,16 @@ "assert g(3.0, 2).numpy() == 6.0 # And its gradient will be 2 * x\n", "assert f(4.0, 3).numpy() == 64.0 # f(x, 3) is essentially x * x * x\n", "assert g(4.0, 3).numpy() == 48.0 # And its gradient will be 3 * x * x" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "aNmR5-jhpX2t" + "id": "aNmR5-jhpX2t", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "At times it may be inconvenient to encapsulate computation of interest into a function. For example, if you want the gradient of the output with respect to intermediate values computed in the function. In such cases, the slightly more verbose but explicit [tf.GradientTape](https://www.tensorflow.org/api_docs/python/tf/GradientTape) context is useful. All computation inside the context of a `tf.GradientTape` is \"recorded\".\n", "\n", @@ -215,19 +289,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "bAFeIE8EuVIq", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "bAFeIE8EuVIq" + } }, - "outputs": [], + "cell_type": "code", "source": [ "x = tf.ones((2, 2))\n", " \n", @@ -249,14 +321,16 @@ "for i in [0, 1]:\n", " for j in [0, 1]:\n", " assert dz_dx[i][j].numpy() == 8.0" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "DK05KXrAAld3" + "id": "DK05KXrAAld3", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Higher-order gradients\n", "\n", @@ -264,19 +338,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "cPQgthZ7ugRJ", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "cPQgthZ7ugRJ" + } }, - "outputs": [], + "cell_type": "code", "source": [ "# TODO(ashankar): Should we use the persistent tape here instead? Follow up on Tom and Alex's discussion\n", "\n", @@ -293,31 +365,21 @@ "\n", "assert dy_dx.numpy() == 3.0\n", "assert d2y_dx2.numpy() == 6.0" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "4U1KKzUpNl58" + "id": "4U1KKzUpNl58", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Next Steps\n", "\n", "In this tutorial we covered gradient computation in TensorFlow. With that we have enough of the primitives required to build an train neural networks, which we will cover in the [next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/3_neural_networks.ipynb)." ] } - ], - "metadata": { - "colab": { - "collapsed_sections": [], - "default_view": {}, - "name": "Automatic Differentiation", - "provenance": [], - "version": "0.3.2", - "views": {} - } - }, - "nbformat": 4, - "nbformat_minor": 0 -} + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb index d268cbcd91..37f2f4f402 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb @@ -1,14 +1,88 @@ { + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Datasets", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, "cells": [ { + "metadata": { + "id": "hvpDiXDuuNEu", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { "metadata": { - "colab_type": "text", - "id": "U9i2Dsh-ziXr" + "id": "4FWzUGc7uQKr", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "RRbeuncot80C", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Datasets" + ] + }, + { + "metadata": { + "id": "i9rtB6DnuFw5", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", "source": [ - "# Eager Execution Tutorial: Importing Data\n", - "\n", "This notebook demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your program. It covers:\n", "\n", "* Creating a `Dataset`.\n", @@ -22,65 +96,53 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "z1JcS5iBXMRO" - }, - "source": [ - "# Setup: Enable eager execution\n" - ] - }, - { - "cell_type": "code", - "execution_count": 0, - "metadata": { - "cellView": "code", + "id": "RlIWhyeLoYnG", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } }, - "colab_type": "code", - "id": "RlIWhyeLoYnG" + "cellView": "code" }, - "outputs": [], + "cell_type": "code", "source": [ "# Import TensorFlow.\n", "import tensorflow as tf\n", "\n", "# Enable eager execution\n", "tf.enable_eager_execution()" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "H9UySOPLXdaw" + "id": "H9UySOPLXdaw", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ - "# Step 1: Create a source `Dataset`\n", + "## Create a source `Dataset`\n", "\n", "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { - "cellView": "code", + "id": "WPTUfGq6kJ5w", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } }, - "colab_type": "code", - "id": "WPTUfGq6kJ5w" + "cellView": "code" }, - "outputs": [], + "cell_type": "code", "source": [ "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", "\n", @@ -93,57 +155,59 @@ "Line 3\n", " \"\"\")\n", "ds_file = tf.data.TextLineDataset(filename)\n" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "twBfWd5xyu_d" + "id": "twBfWd5xyu_d", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ - "# Step 2: Apply transformations\n", + "## Apply transformations\n", "\n", "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { - "cellView": "code", + "id": "ngUe237Wt48W", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } }, - "colab_type": "code", - "id": "ngUe237Wt48W" + "cellView": "code" }, - "outputs": [], + "cell_type": "code", "source": [ "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", "ds_file = ds_file.batch(2)" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "IDY4WsYRhP81" + "id": "IDY4WsYRhP81", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ - "# Step 3: Iterate\n", + "## Iterate\n", "\n", "When eager execution is enabled `Dataset` objects support iteration.\n", "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -152,7 +216,7 @@ "base_uri": "https://localhost:8080/", "height": 153 }, - "colab_type": "code", + "outputId": "8e4b0298-d27d-4ac7-e26a-ef94af0594ec", "executionInfo": { "elapsed": 388, "status": "ok", @@ -163,13 +227,21 @@ "userId": "" }, "user_tz": 420 - }, - "id": "lCUWzso6mbqR", - "outputId": "8e4b0298-d27d-4ac7-e26a-ef94af0594ec" + } }, + "cell_type": "code", + "source": [ + "print('Elements of ds_tensors:')\n", + "for x in ds_tensors:\n", + " print(x)\n", + "\n", + "print('\\nElements in ds_file:')\n", + "for x in ds_file:\n", + " print(x)" + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "Elements of ds_tensors:\n", @@ -180,30 +252,10 @@ "Elements in ds_file:\n", "tf.Tensor(['Line 1' 'Line 2'], shape=(2,), dtype=string)\n", "tf.Tensor(['Line 3' ' '], shape=(2,), dtype=string)\n" - ] + ], + "name": "stdout" } - ], - "source": [ - "print('Elements of ds_tensors:')\n", - "for x in ds_tensors:\n", - " print(x)\n", - "\n", - "print('\\nElements in ds_file:')\n", - "for x in ds_file:\n", - " print(x)" ] } - ], - "metadata": { - "colab": { - "collapsed_sections": [], - "default_view": {}, - "name": "Eager Execution Tutorial: Importing Data", - "provenance": [], - "version": "0.3.2", - "views": {} - } - }, - "nbformat": 4, - "nbformat_minor": 0 -} + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb index 84f1d031d4..ca0262eade 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb @@ -1,14 +1,88 @@ { + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Variables, models, and training", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, "cells": [ { + "metadata": { + "id": "5rmpybwysXGV", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { "metadata": { - "colab_type": "text", - "id": "k2o3TTG4TFpt" + "id": "m8y3rGtQsYP2", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "hrXv0rU9sIma", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Variables, models, and training" + ] + }, + { + "metadata": { + "id": "7S0BwJ_8sLu7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "k2o3TTG4TFpt", + "colab_type": "text" + }, + "cell_type": "markdown", "source": [ - "# Training Models\n", - "\n", "In the previous tutorial we covered the TensorFlow APIs for automatic differentiation, a basic building block for machine learning.\n", "In this tutorial we will use the TensorFlow primitives introduced in the prior tutorials to do some simple machine learning.\n", "\n", @@ -16,41 +90,41 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "3LXMVuV0VhDr" + "id": "3LXMVuV0VhDr", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Setup" ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "PJ64L90aVir3", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "PJ64L90aVir3" + } }, - "outputs": [], + "cell_type": "code", "source": [ "import tensorflow as tf\n", "tf.enable_eager_execution()\n", "tfe = tf.contrib.eager # Shorthand for some symbols" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "eMAWbDJFVmMk" + "id": "eMAWbDJFVmMk", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Variables\n", "\n", @@ -58,33 +132,33 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "VkJwtLS_Jbn8", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "VkJwtLS_Jbn8" + } }, - "outputs": [], + "cell_type": "code", "source": [ "# Using python state\n", "x = tf.zeros([10, 10])\n", "x += 2 # This is equivalent to x = x + 2, which does not mutate the original\n", " # value of x\n", "print(x)" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "wfneTXy7JcUz" + "id": "wfneTXy7JcUz", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "TensorFlow, however, has stateful operations built in, and these are often more pleasant to use than low-level Python representations of your state. To represent weights in a model, for example, it's often convenient and efficient to use TensorFlow variables.\n", "\n", @@ -92,19 +166,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "itxmrMil6DQi", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "itxmrMil6DQi" + } }, - "outputs": [], + "cell_type": "code", "source": [ "v = tfe.Variable(1.0)\n", "assert v.numpy() == 1.0\n", @@ -116,14 +188,16 @@ "# Use `v` in a TensorFlow operation like tf.square() and reassign\n", "v.assign(tf.square(v))\n", "assert v.numpy() == 9.0" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "-paSaeq1JzwC" + "id": "-paSaeq1JzwC", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Computations using Variables are automatically traced when computing gradients. For Variables representing embeddings TensorFlow will do sparse updates by default, which are more computation and memory efficient.\n", "\n", @@ -131,11 +205,11 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "BMiFcDzE7Qu3" + "id": "BMiFcDzE7Qu3", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Example: Fitting a linear model\n", "\n", @@ -150,11 +224,11 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "gFzH64Jn9PIm" + "id": "gFzH64Jn9PIm", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Define the model\n", "\n", @@ -162,19 +236,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "_WRu7Pze7wk8", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "_WRu7Pze7wk8" + } }, - "outputs": [], + "cell_type": "code", "source": [ "class Model(object):\n", " def __init__(self):\n", @@ -189,14 +261,16 @@ "model = Model()\n", "\n", "assert model(3.0).numpy() == 15.0" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "xa6j_yXa-j79" + "id": "xa6j_yXa-j79", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Define a loss function\n", "\n", @@ -204,30 +278,30 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "Y0ysUFGY924U", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "Y0ysUFGY924U" + } }, - "outputs": [], + "cell_type": "code", "source": [ "def loss(predicted_y, desired_y):\n", " return tf.reduce_mean(tf.square(predicted_y - desired_y))" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "qutT_fkl_CBc" + "id": "qutT_fkl_CBc", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Obtain training data\n", "\n", @@ -235,19 +309,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "gxPTb-kt_N5m", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "gxPTb-kt_N5m" + } }, - "outputs": [], + "cell_type": "code", "source": [ "TRUE_W = 3.0\n", "TRUE_b = 2.0\n", @@ -256,22 +328,24 @@ "inputs = tf.random_normal(shape=[NUM_EXAMPLES])\n", "noise = tf.random_normal(shape=[NUM_EXAMPLES])\n", "outputs = inputs * TRUE_W + TRUE_b + noise" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "-50nq-wPBsAW" + "id": "-50nq-wPBsAW", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Before we train the model let's visualize where the model stands right now. We'll plot the model's predictions in red and the training data in blue." ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "_eb83LtrB4nt", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -279,7 +353,7 @@ }, "height": 293 }, - "colab_type": "code", + "outputId": "3873f508-72fb-41e7-a7f5-3f513deefe38", "executionInfo": { "elapsed": 1210, "status": "ok", @@ -290,48 +364,48 @@ "userId": "" }, "user_tz": 420 - }, - "id": "_eb83LtrB4nt", - "outputId": "3873f508-72fb-41e7-a7f5-3f513deefe38" + } }, + "cell_type": "code", + "source": [ + "import matplotlib.pyplot as plt\n", + "\n", + "plt.scatter(inputs, outputs, c='b')\n", + "plt.scatter(inputs, model(inputs), c='r')\n", + "plt.show()\n", + "\n", + "print('Current loss: '),\n", + "print(loss(model(inputs), outputs).numpy())" + ], + "execution_count": 0, "outputs": [ { + "output_type": "display_data", "data": { "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEDCAYAAAA2k7/eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\noZkggi8IgtBMEMEXBEFoJvw//5K32R/vBHAAAAAASUVORK5CYII=\n", "text/plain": [ - "\u003cmatplotlib.figure.Figure at 0x7f5be3c99f50\u003e" + "" ] }, "metadata": { "tags": [] - }, - "output_type": "display_data" + } }, { - "name": "stdout", "output_type": "stream", "text": [ "Current loss: 9.48636\n" - ] + ], + "name": "stdout" } - ], - "source": [ - "import matplotlib.pyplot as plt\n", - "\n", - "plt.scatter(inputs, outputs, c='b')\n", - "plt.scatter(inputs, model(inputs), c='r')\n", - "plt.show()\n", - "\n", - "print('Current loss: '),\n", - "print(loss(model(inputs), outputs).numpy())" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "sSDP-yeq_4jE" + "id": "sSDP-yeq_4jE", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "### Define a training loop\n", "\n", @@ -339,19 +413,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "MBIACgdnA55X", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "MBIACgdnA55X" + } }, - "outputs": [], + "cell_type": "code", "source": [ "def train(model, inputs, outputs, learning_rate):\n", " with tf.GradientTape() as t:\n", @@ -359,22 +431,24 @@ " dW, db = t.gradient(current_loss, [model.W, model.b])\n", " model.W.assign_sub(learning_rate * dW)\n", " model.b.assign_sub(learning_rate * db)" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "RwWPaJryD2aN" + "id": "RwWPaJryD2aN", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Finally, let's repeatedly run through the training data and see how `W` and `b` evolve." ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "XdfkR223D9dW", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -382,7 +456,7 @@ }, "height": 446 }, - "colab_type": "code", + "outputId": "c43591ae-d5ac-4f2b-a8e7-bfce607e0919", "executionInfo": { "elapsed": 569, "status": "ok", @@ -393,13 +467,36 @@ "userId": "" }, "user_tz": 420 - }, - "id": "XdfkR223D9dW", - "outputId": "c43591ae-d5ac-4f2b-a8e7-bfce607e0919" + } }, + "cell_type": "code", + "source": [ + "model = Model()\n", + "\n", + "# Collect the history of W-values and b-values to plot later\n", + "Ws, bs = [], []\n", + "epochs = range(10)\n", + "for epoch in epochs:\n", + " Ws.append(model.W.numpy())\n", + " bs.append(model.b.numpy())\n", + " current_loss = loss(model(inputs), outputs)\n", + "\n", + " train(model, inputs, outputs, learning_rate=0.1)\n", + " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", + " (epoch, Ws[-1], bs[-1], current_loss))\n", + "\n", + "# Let's plot it all\n", + "plt.plot(epochs, Ws, 'r',\n", + " epochs, bs, 'b')\n", + "plt.plot([TRUE_W] * len(epochs), 'r--',\n", + " [TRUE_b] * len(epochs), 'b--')\n", + "plt.legend(['W', 'b', 'true W', 'true_b'])\n", + "plt.show()\n", + " " + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "Epoch 0: W=5.00 b=0.00, loss=9.48636\n", @@ -412,52 +509,29 @@ "Epoch 7: W=3.38 b=1.62, loss=1.34756\n", "Epoch 8: W=3.30 b=1.70, loss=1.23463\n", "Epoch 9: W=3.24 b=1.76, loss=1.16460\n" - ] + ], + "name": "stdout" }, { + "output_type": "display_data", "data": { "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW0AAAEDCAYAAAD+/1UIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n", "text/plain": [ - "\u003cmatplotlib.figure.Figure at 0x7f5be4b8ec50\u003e" + "" ] }, "metadata": { "tags": [] - }, - "output_type": "display_data" + } } - ], - "source": [ - "model = Model()\n", - "\n", - "# Collect the history of W-values and b-values to plot later\n", - "Ws, bs = [], []\n", - "epochs = range(10)\n", - "for epoch in epochs:\n", - " Ws.append(model.W.numpy())\n", - " bs.append(model.b.numpy())\n", - " current_loss = loss(model(inputs), outputs)\n", - "\n", - " train(model, inputs, outputs, learning_rate=0.1)\n", - " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", - " (epoch, Ws[-1], bs[-1], current_loss))\n", - "\n", - "# Let's plot it all\n", - "plt.plot(epochs, Ws, 'r',\n", - " epochs, bs, 'b')\n", - "plt.plot([TRUE_W] * len(epochs), 'r--',\n", - " [TRUE_b] * len(epochs), 'b--')\n", - "plt.legend(['W', 'b', 'true W', 'true_b'])\n", - "plt.show()\n", - " " ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "vPnIVuaSJwWz" + "id": "vPnIVuaSJwWz", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Next Steps\n", "\n", @@ -469,17 +543,5 @@ "The [next tutorial](TODO) will cover these higher level APIs." ] } - ], - "metadata": { - "colab": { - "collapsed_sections": [], - "default_view": {}, - "name": "Training Models", - "provenance": [], - "version": "0.3.2", - "views": {} - } - }, - "nbformat": 4, - "nbformat_minor": 0 -} + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb index 5749f22ac5..3378054be4 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb @@ -1,44 +1,122 @@ { + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Custom layers", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + }, + "kernelspec": { + "display_name": "Python 3", + "name": "python3" + } + }, "cells": [ { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "tDnwEv8FtJm7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "JlknJBWQtKkI", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } }, - "colab_type": "code", - "id": "pwX7Fii1rwsJ" + "cellView": "form" }, - "outputs": [], + "cell_type": "code", "source": [ - "import tensorflow as tf\n", - "tf.enable_eager_execution()\n", - "tfe = tf.contrib.eager\n" - ] + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] }, { + "metadata": { + "id": "60RdWsg1tETW", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "# Custom layers" + ] + }, + { "metadata": { - "colab_type": "text", - "id": "UEu3q4jmpKVT" + "id": "BcJg7Enms86w", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ - "# High level API\n", - "\n", - "We recommend using `tf.keras` as a high-level API for building neural networks. That said, most TensorFlow APIs are usable with eager execution.\n", - "\n" + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" ] }, { + "metadata": { + "id": "UEu3q4jmpKVT", + "colab_type": "text" + }, "cell_type": "markdown", + "source": [ + "We recommend using `tf.keras` as a high-level API for building neural networks. That said, most TensorFlow APIs are usable with eager execution.\n" + ] + }, + { "metadata": { - "colab_type": "text", - "id": "zSFfVVjkrrsI" + "id": "pwX7Fii1rwsJ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "tfe = tf.contrib.eager\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "zSFfVVjkrrsI", + "colab_type": "text" + }, + "cell_type": "markdown", "source": [ "## Layers: common sets of useful operations\n", "\n", @@ -50,19 +128,17 @@ ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "8PyXlPl-4TzQ", + "colab_type": "code", "colab": { "autoexec": { "startup": false, "wait_interval": 0 } - }, - "colab_type": "code", - "id": "8PyXlPl-4TzQ" + } }, - "outputs": [], + "cell_type": "code", "source": [ "# In the tf.keras.layers package, layers are objects. To construct a layer,\n", "# simply construct the object. Most layers take as a first argument the number\n", @@ -72,23 +148,25 @@ "# the first time the layer is used, but it can be provided if you want to \n", "# specify it manually, which is useful in some complex models.\n", "layer = tf.keras.layers.Dense(10, input_shape=(None, 5))" - ] + ], + "execution_count": 0, + "outputs": [] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "Fn69xxPO5Psr" + "id": "Fn69xxPO5Psr", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "The full list of pre-existing layers can be seen in [the documentation](https://www.tensorflow.org/api_docs/python/tf/keras/layers). It includes Dense (a fully-connected layer),\n", "Conv2D, LSTM, BatchNormalization, Dropout, and many others." ] }, { - "cell_type": "code", - "execution_count": 3, "metadata": { + "id": "E3XKNknP5Mhb", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -96,7 +174,7 @@ }, "height": 204 }, - "colab_type": "code", + "outputId": "c5d52434-d980-4488-efa7-5660819d0207", "executionInfo": { "elapsed": 244, "status": "ok", @@ -107,15 +185,20 @@ "userId": "" }, "user_tz": 420 - }, - "id": "E3XKNknP5Mhb", - "outputId": "c5d52434-d980-4488-efa7-5660819d0207" + } }, + "cell_type": "code", + "source": [ + "# To use a layer, simply call it.\n", + "layer(tf.zeros([10, 5]))" + ], + "execution_count": 0, "outputs": [ { + "output_type": "execute_result", "data": { "text/plain": [ - "\u003ctf.Tensor: id=30, shape=(10, 10), dtype=float32, numpy=\n", + "" ] }, - "execution_count": 3, "metadata": { "tags": [] }, - "output_type": "execute_result" + "execution_count": 3 } - ], - "source": [ - "# To use a layer, simply call it.\n", - "layer(tf.zeros([10, 5]))" ] }, { - "cell_type": "code", - "execution_count": 4, "metadata": { + "id": "Wt_Nsv-L5t2s", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -151,7 +229,7 @@ }, "height": 221 }, - "colab_type": "code", + "outputId": "f0d96dce-0128-4080-bfe2-0ee6fbc0ad90", "executionInfo": { "elapsed": 320, "status": "ok", @@ -162,15 +240,22 @@ "userId": "" }, "user_tz": 420 - }, - "id": "Wt_Nsv-L5t2s", - "outputId": "f0d96dce-0128-4080-bfe2-0ee6fbc0ad90" + } }, + "cell_type": "code", + "source": [ + "# Layers have many useful methods. For example, you can inspect all variables\n", + "# in a layer by calling layer.variables. In this case a fully-connected layer\n", + "# will have variables for weights and biases.\n", + "layer.variables" + ], + "execution_count": 0, "outputs": [ { + "output_type": "execute_result", "data": { "text/plain": [ - "[\u003ctf.Variable 'dense_1/kernel:0' shape=(5, 10) dtype=float32, numpy=\n", + "[,\n", + " ]" ] }, - "execution_count": 4, "metadata": { "tags": [] }, - "output_type": "execute_result" + "execution_count": 4 } - ], - "source": [ - "# Layers have many useful methods. For example, you can inspect all variables\n", - "# in a layer by calling layer.variables. In this case a fully-connected layer\n", - "# will have variables for weights and biases.\n", - "layer.variables" ] }, { - "cell_type": "code", - "execution_count": 5, "metadata": { + "id": "6ilvKjz8_4MQ", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -209,7 +287,7 @@ }, "height": 221 }, - "colab_type": "code", + "outputId": "f647fced-c2d7-41a3-c237-242036784665", "executionInfo": { "elapsed": 226, "status": "ok", @@ -220,15 +298,20 @@ "userId": "" }, "user_tz": 420 - }, - "id": "6ilvKjz8_4MQ", - "outputId": "f647fced-c2d7-41a3-c237-242036784665" + } }, + "cell_type": "code", + "source": [ + "# The variables are also accessible through nice accessors\n", + "layer.kernel, layer.bias" + ], + "execution_count": 0, "outputs": [ { + "output_type": "execute_result", "data": { "text/plain": [ - "(\u003ctf.Variable 'dense_1/kernel:0' shape=(5, 10) dtype=float32, numpy=\n", + "(,\n", + " )" ] }, - "execution_count": 5, "metadata": { "tags": [] }, - "output_type": "execute_result" + "execution_count": 5 } - ], - "source": [ - "# The variables are also accessible through nice accessors\n", - "layer.kernel, layer.bias" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "O0kDbE54-5VS" + "id": "O0kDbE54-5VS", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Implementing custom layers\n", "The best way to implement your own layer is extending the tf.keras.Layer class and implementing:\n", @@ -271,9 +349,9 @@ ] }, { - "cell_type": "code", - "execution_count": 7, "metadata": { + "id": "5Byl3n1k5kIy", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -281,7 +359,7 @@ }, "height": 391 }, - "colab_type": "code", + "outputId": "6e7f9285-649a-4132-82ce-73ea92f15862", "executionInfo": { "elapsed": 251, "status": "ok", @@ -292,13 +370,30 @@ "userId": "" }, "user_tz": 420 - }, - "id": "5Byl3n1k5kIy", - "outputId": "6e7f9285-649a-4132-82ce-73ea92f15862" + } }, + "cell_type": "code", + "source": [ + "class MyDenseLayer(tf.keras.layers.Layer):\n", + " def __init__(self, num_outputs):\n", + " super(MyDenseLayer, self).__init__()\n", + " self.num_outputs = num_outputs\n", + " \n", + " def build(self, input_shape):\n", + " self.kernel = self.add_variable(\"kernel\", \n", + " shape=[input_shape[-1].value, \n", + " self.num_outputs])\n", + " \n", + " def call(self, input):\n", + " return tf.matmul(input, self.kernel)\n", + " \n", + "layer = MyDenseLayer(10)\n", + "print(layer(tf.zeros([10, 5])))\n", + "print(layer.variables)" + ], + "execution_count": 0, "outputs": [ { - "name": "stdout", "output_type": "stream", "text": [ "tf.Tensor(\n", @@ -312,7 +407,7 @@ " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(10, 10), dtype=float32)\n", - "[\u003ctf.Variable 'my_dense_layer_1/kernel:0' shape=(5, 10) dtype=float32, numpy=\n", + "[]\n" + ], + "name": "stdout" } - ], - "source": [ - "class MyDenseLayer(tf.keras.layers.Layer):\n", - " def __init__(self, num_outputs):\n", - " super(MyDenseLayer, self).__init__()\n", - " self.num_outputs = num_outputs\n", - " \n", - " def build(self, input_shape):\n", - " self.kernel = self.add_variable(\"kernel\", \n", - " shape=[input_shape[-1].value, \n", - " self.num_outputs])\n", - " \n", - " def call(self, input):\n", - " return tf.matmul(input, self.kernel)\n", - " \n", - "layer = MyDenseLayer(10)\n", - "print(layer(tf.zeros([10, 5])))\n", - "print(layer.variables)" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "tk8E2vY0-z4Z" + "id": "tk8E2vY0-z4Z", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Note that you don't have to wait until `build` is called to create your variables, you can also create them in `__init__`.\n", "\n", @@ -358,11 +436,11 @@ ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "Qhg4KlbKrs3G" + "id": "Qhg4KlbKrs3G", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "## Models: composing layers\n", "\n", @@ -372,9 +450,9 @@ ] }, { - "cell_type": "code", - "execution_count": 9, "metadata": { + "id": "N30DTXiRASlb", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -382,7 +460,7 @@ }, "height": 190 }, - "colab_type": "code", + "outputId": "a8b23a8e-5cf9-4bbf-f93b-6c763d74e2b3", "executionInfo": { "elapsed": 420, "status": "ok", @@ -393,27 +471,9 @@ "userId": "" }, "user_tz": 420 - }, - "id": "N30DTXiRASlb", - "outputId": "a8b23a8e-5cf9-4bbf-f93b-6c763d74e2b3" - }, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "tf.Tensor(\n", - "[[[[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]\n", - "\n", - " [[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]]], shape=(1, 2, 3, 3), dtype=float32)\n", - "['resnet_identity_block_1/conv2d_3/kernel:0', 'resnet_identity_block_1/conv2d_3/bias:0', 'resnet_identity_block_1/batch_normalization_3/gamma:0', 'resnet_identity_block_1/batch_normalization_3/beta:0', 'resnet_identity_block_1/conv2d_4/kernel:0', 'resnet_identity_block_1/conv2d_4/bias:0', 'resnet_identity_block_1/batch_normalization_4/gamma:0', 'resnet_identity_block_1/batch_normalization_4/beta:0', 'resnet_identity_block_1/conv2d_5/kernel:0', 'resnet_identity_block_1/conv2d_5/bias:0', 'resnet_identity_block_1/batch_normalization_5/gamma:0', 'resnet_identity_block_1/batch_normalization_5/beta:0', 'resnet_identity_block_1/batch_normalization_3/moving_mean:0', 'resnet_identity_block_1/batch_normalization_3/moving_variance:0', 'resnet_identity_block_1/batch_normalization_4/moving_mean:0', 'resnet_identity_block_1/batch_normalization_4/moving_variance:0', 'resnet_identity_block_1/batch_normalization_5/moving_mean:0', 'resnet_identity_block_1/batch_normalization_5/moving_variance:0']\n" - ] } - ], + }, + "cell_type": "code", "source": [ "class ResnetIdentityBlock(tf.keras.Model):\n", " def __init__(self, kernel_size, filters):\n", @@ -448,22 +508,40 @@ "block = ResnetIdentityBlock(1, [1, 2, 3])\n", "print(block(tf.zeros([1, 2, 3, 3])))\n", "print([x.name for x in block.variables])" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "tf.Tensor(\n", + "[[[[ 0. 0. 0.]\n", + " [ 0. 0. 0.]\n", + " [ 0. 0. 0.]]\n", + "\n", + " [[ 0. 0. 0.]\n", + " [ 0. 0. 0.]\n", + " [ 0. 0. 0.]]]], shape=(1, 2, 3, 3), dtype=float32)\n", + "['resnet_identity_block_1/conv2d_3/kernel:0', 'resnet_identity_block_1/conv2d_3/bias:0', 'resnet_identity_block_1/batch_normalization_3/gamma:0', 'resnet_identity_block_1/batch_normalization_3/beta:0', 'resnet_identity_block_1/conv2d_4/kernel:0', 'resnet_identity_block_1/conv2d_4/bias:0', 'resnet_identity_block_1/batch_normalization_4/gamma:0', 'resnet_identity_block_1/batch_normalization_4/beta:0', 'resnet_identity_block_1/conv2d_5/kernel:0', 'resnet_identity_block_1/conv2d_5/bias:0', 'resnet_identity_block_1/batch_normalization_5/gamma:0', 'resnet_identity_block_1/batch_normalization_5/beta:0', 'resnet_identity_block_1/batch_normalization_3/moving_mean:0', 'resnet_identity_block_1/batch_normalization_3/moving_variance:0', 'resnet_identity_block_1/batch_normalization_4/moving_mean:0', 'resnet_identity_block_1/batch_normalization_4/moving_variance:0', 'resnet_identity_block_1/batch_normalization_5/moving_mean:0', 'resnet_identity_block_1/batch_normalization_5/moving_variance:0']\n" + ], + "name": "stdout" + } ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "wYfucVw65PMj" + "id": "wYfucVw65PMj", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "Much of the time, however, models which compose many layers simply call one layer after the other. This can be done in very little code using tf.keras.Sequential" ] }, { - "cell_type": "code", - "execution_count": 0, "metadata": { + "id": "L9frk7Ur4uvJ", + "colab_type": "code", "colab": { "autoexec": { "startup": false, @@ -472,7 +550,7 @@ "base_uri": "https://localhost:8080/", "height": 153 }, - "colab_type": "code", + "outputId": "882e9076-b6d9-4380-bb1e-7c6b57d54c39", "executionInfo": { "elapsed": 361, "status": "ok", @@ -483,69 +561,53 @@ "userId": "108023195365833072773" }, "user_tz": 420 - }, - "id": "L9frk7Ur4uvJ", - "outputId": "882e9076-b6d9-4380-bb1e-7c6b57d54c39" + } }, + "cell_type": "code", + "source": [ + " my_seq = tf.keras.Sequential([tf.keras.layers.Conv2D(1, (1, 1)),\n", + " tf.keras.layers.BatchNormalization(),\n", + " tf.keras.layers.Conv2D(2, 1, \n", + " padding='same'),\n", + " tf.keras.layers.BatchNormalization(),\n", + " tf.keras.layers.Conv2D(3, (1, 1)),\n", + " tf.keras.layers.BatchNormalization()])\n", + "my_seq(tf.zeros([1, 2, 3, 3]))" + ], + "execution_count": 0, "outputs": [ { + "output_type": "execute_result", "data": { "text/plain": [ - "\u003ctf.Tensor: id=1423, shape=(1, 2, 3, 3), dtype=float32, numpy=\n", + "" ] }, - "execution_count": 26, "metadata": { "tags": [] }, - "output_type": "execute_result" + "execution_count": 26 } - ], - "source": [ - " my_seq = tf.keras.Sequential([tf.keras.layers.Conv2D(1, (1, 1)),\n", - " tf.keras.layers.BatchNormalization(),\n", - " tf.keras.layers.Conv2D(2, 1, \n", - " padding='same'),\n", - " tf.keras.layers.BatchNormalization(),\n", - " tf.keras.layers.Conv2D(3, (1, 1)),\n", - " tf.keras.layers.BatchNormalization()])\n", - "my_seq(tf.zeros([1, 2, 3, 3]))" ] }, { - "cell_type": "markdown", "metadata": { - "colab_type": "text", - "id": "c5YwYcnuK-wc" + "id": "c5YwYcnuK-wc", + "colab_type": "text" }, + "cell_type": "markdown", "source": [ "# Next steps\n", "\n", "Now you can go back to the previous notebook and adapt the linear regression example to use layers and models to be better structured." ] } - ], - "metadata": { - "colab": { - "collapsed_sections": [], - "default_view": {}, - "name": "4 - High level API - TensorFlow Eager.ipynb", - "provenance": [], - "version": "0.3.2", - "views": {} - }, - "kernelspec": { - "display_name": "Python 3", - "name": "python3" - } - }, - "nbformat": 4, - "nbformat_minor": 0 -} + ] +} \ No newline at end of file -- cgit v1.2.3 From 1f72a63d1ae27544ca9610e4de0080a5e121a62f Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 16:43:21 -0700 Subject: Rename eager notebook filenames to match getting started collection --- .../eager/python/examples/notebooks/1_basics.ipynb | 491 ----------------- .../python/examples/notebooks/2_gradients.ipynb | 385 ------------- .../python/examples/notebooks/3_datasets.ipynb | 261 --------- .../examples/notebooks/3_training_models.ipynb | 547 ------------------ .../python/examples/notebooks/4_high_level.ipynb | 613 --------------------- .../eager/python/examples/notebooks/README.md | 11 + .../notebooks/automatic_differentiation.ipynb | 385 +++++++++++++ .../python/examples/notebooks/custom_layers.ipynb | 613 +++++++++++++++++++++ .../eager/python/examples/notebooks/datasets.ipynb | 261 +++++++++ .../python/examples/notebooks/eager_basics.ipynb | 491 +++++++++++++++++ .../notebooks/variables_models_training.ipynb | 547 ++++++++++++++++++ 11 files changed, 2308 insertions(+), 2297 deletions(-) delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/README.md create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb diff --git a/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb deleted file mode 100644 index 99d141bf2b..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/1_basics.ipynb +++ /dev/null @@ -1,491 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Eager execution basics", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "collapsed_sections": [], - "toc_visible": true - } - }, - "cells": [ - { - "metadata": { - "id": "iPpI7RaYoZuE", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "hro2InpHobKk", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "U9i2Dsh-ziXr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Eager execution basics" - ] - }, - { - "metadata": { - "id": "Hndw-YcxoOJK", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
\n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
" - ] - }, - { - "metadata": { - "id": "6sILUVbHoSgH", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "This is an introductory tutorial for using TensorFlow. It will cover:\n", - "\n", - "* Importing required packages\n", - "* Creating and using Tensors\n", - "* Using GPU acceleration" - ] - }, - { - "metadata": { - "id": "z1JcS5iBXMRO", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Import TensorFlow\n", - "\n", - "To get started, import the `tensorflow` module and enable eager execution.\n", - "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." - ] - }, - { - "metadata": { - "id": "RlIWhyeLoYnG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "H9UySOPLXdaw", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Tensors\n", - "\n", - "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" - ] - }, - { - "metadata": { - "id": "ngUe237Wt48W", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 125 - }, - "cellView": "code", - "outputId": "b1a1cd60-4eb3-443d-cd6b-68406390784e", - "executionInfo": { - "elapsed": 320, - "status": "ok", - "timestamp": 1526420535530, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "print(tf.add(1, 2))\n", - "print(tf.add([1, 2], [3, 4]))\n", - "print(tf.square(5))\n", - "print(tf.reduce_sum([1, 2, 3]))\n", - "print(tf.encode_base64(\"hello world\"))\n", - "\n", - "# Operator overloading is also supported\n", - "print(tf.square(2) + tf.square(3))" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(3, shape=(), dtype=int32)\n", - "tf.Tensor([4 6], shape=(2,), dtype=int32)\n", - "tf.Tensor(25, shape=(), dtype=int32)\n", - "tf.Tensor(6, shape=(), dtype=int32)\n", - "tf.Tensor(aGVsbG8gd29ybGQ, shape=(), dtype=string)\n", - "tf.Tensor(13, shape=(), dtype=int32)\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "IDY4WsYRhP81", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Each Tensor has a shape and a datatype" - ] - }, - { - "metadata": { - "id": "srYWH1MdJNG7", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 53 - }, - "outputId": "5e4ac41c-5115-4e50-eba0-42e249c16561", - "executionInfo": { - "elapsed": 215, - "status": "ok", - "timestamp": 1526420538162, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "x = tf.matmul([[1]], [[2, 3]])\n", - "print(x.shape)\n", - "print(x.dtype)" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "(1, 2)\n", - "\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "eBPw8e8vrsom", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", - "\n", - "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", - "2. Tensors are immutable." - ] - }, - { - "metadata": { - "id": "Dwi1tdW3JBw6", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### NumPy Compatibility\n", - "\n", - "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", - "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", - "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", - "\n", - "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", - "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." - ] - }, - { - "metadata": { - "id": "lCUWzso6mbqR", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 251 - }, - "outputId": "fd0a22bc-8249-49dd-fcbd-63161cc47e46", - "executionInfo": { - "elapsed": 238, - "status": "ok", - "timestamp": 1526420540562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "import numpy as np\n", - "\n", - "ndarray = np.ones([3, 3])\n", - "\n", - "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", - "tensor = tf.multiply(ndarray, 42)\n", - "print(tensor)\n", - "\n", - "\n", - "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", - "print(np.add(tensor, 1))\n", - "\n", - "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", - "print(tensor.numpy())" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "TensorFlow operations convert numpy arrays to Tensors automatically\n", - "tf.Tensor(\n", - "[[ 42. 42. 42.]\n", - " [ 42. 42. 42.]\n", - " [ 42. 42. 42.]], shape=(3, 3), dtype=float64)\n", - "And NumPy operations convert Tensors to numpy arrays automatically\n", - "[[ 43. 43. 43.]\n", - " [ 43. 43. 43.]\n", - " [ 43. 43. 43.]]\n", - "The .numpy() method explicitly converts a Tensor to a numpy array\n", - "[[ 42. 42. 42.]\n", - " [ 42. 42. 42.]\n", - " [ 42. 42. 42.]]\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "PBNP8yTRfu_X", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## GPU acceleration\n", - "\n", - "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" - ] - }, - { - "metadata": { - "id": "3Twf_Rw-gQFM", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 53 - }, - "cellView": "code", - "outputId": "2239ae2b-adf3-4895-b1f3-464cf5361d1b", - "executionInfo": { - "elapsed": 340, - "status": "ok", - "timestamp": 1526420543562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "x = tf.random_uniform([3, 3])\n", - "\n", - "print(\"Is there a GPU available: \"),\n", - "print(tf.test.is_gpu_available())\n", - "\n", - "print(\"Is the Tensor on GPU #0: \"),\n", - "print(x.device.endswith('GPU:0'))" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Is there a GPU available: False\n", - "Is the Tensor on GPU #0: False\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "vpgYzgVXW2Ud", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Device Names\n", - "\n", - "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." - ] - }, - { - "metadata": { - "id": "ZWZQCimzuqyP", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "\n", - "\n", - "### Explicit Device Placement\n", - "\n", - "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" - ] - }, - { - "metadata": { - "id": "RjkNZTuauy-Q", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 53 - }, - "outputId": "2e613293-ccac-4db2-b793-8ceb5b5adcfd", - "executionInfo": { - "elapsed": 1762, - "status": "ok", - "timestamp": 1526420547562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "def time_matmul(x):\n", - " %timeit tf.matmul(x, x)\n", - "\n", - "# Force execution on CPU\n", - "print(\"On CPU:\")\n", - "with tf.device(\"CPU:0\"):\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"CPU:0\")\n", - " time_matmul(x)\n", - "\n", - "# Force execution on GPU #0 if available\n", - "if tf.test.is_gpu_available():\n", - " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"GPU:0\")\n", - " time_matmul(x)" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "On CPU:\n", - "10 loops, best of 3: 35.8 ms per loop\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "YEOJTNiOvnpQ", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Next Steps\n", - "\n", - "In this tutorial we covered the most fundamental concepts in TensorFlow - `Tensor`s, operations, and devices.\n", - "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb deleted file mode 100644 index d7bcf749e8..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/2_gradients.ipynb +++ /dev/null @@ -1,385 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Automatic differentiation and gradient tape", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "collapsed_sections": [], - "toc_visible": true - } - }, - "cells": [ - { - "metadata": { - "id": "t09eeeR5prIJ", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "GCCk8_dHpuNf", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "xh8WkEwWpnm7", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Automatic differentiation and gradient tape" - ] - }, - { - "metadata": { - "id": "idv0bPeCp325", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
\n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
" - ] - }, - { - "metadata": { - "id": "vDJ4XzMqodTy", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "In the previous tutorial we introduced `Tensor`s and operations on them. In this tutorial we will cover [automatic differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation), a key technique for optimizing machine learning models." - ] - }, - { - "metadata": { - "id": "GQJysDM__Qb0", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Setup\n" - ] - }, - { - "metadata": { - "id": "OiMPZStlibBv", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "tf.enable_eager_execution()\n", - "\n", - "tfe = tf.contrib.eager # Shorthand for some symbols" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "1CLWJl0QliB0", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Derivatives of a function\n", - "\n", - "TensorFlow provides APIs for automatic differentiation - computing the derivative of a function. The way that more closely mimics the math is to encapsulate the computation in a Python function, say `f`, and use `tfe.gradients_function` to create a function that computes the derivatives of `f` with respect to its arguments. If you're familiar with [autograd](https://github.com/HIPS/autograd) for differentiating numpy functions, this will be familiar. For example: " - ] - }, - { - "metadata": { - "id": "9FViq92UX7P8", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "from math import pi\n", - "\n", - "def f(x):\n", - " return tf.square(tf.sin(x))\n", - "\n", - "assert f(pi/2).numpy() == 1.0\n", - "\n", - "\n", - "# grad_f will return a list of derivatives of f\n", - "# with respect to its arguments. Since f() has a single argument,\n", - "# grad_f will return a list with a single element.\n", - "grad_f = tfe.gradients_function(f)\n", - "assert tf.abs(grad_f(pi/2)[0]).numpy() < 1e-7" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "v9fPs8RyopCf", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Higher-order gradients\n", - "\n", - "The same API can be used to differentiate as many times as you like:\n" - ] - }, - { - "metadata": { - "id": "3D0ZvnGYo0rW", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 276 - }, - "outputId": "e23f8cc6-6813-4944-f20f-825b8a03c2ff", - "executionInfo": { - "elapsed": 730, - "status": "ok", - "timestamp": 1527005655565, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "def f(x):\n", - " return tf.square(tf.sin(x))\n", - "\n", - "def grad(f):\n", - " return lambda x: tfe.gradients_function(f)(x)[0]\n", - "\n", - "x = tf.lin_space(-2*pi, 2*pi, 100) # 100 points between -2Ï€ and +2Ï€\n", - "\n", - "import matplotlib.pyplot as plt\n", - "\n", - "plt.plot(x, f(x), label=\"f\")\n", - "plt.plot(x, grad(f)(x), label=\"first derivative\")\n", - "plt.plot(x, grad(grad(f))(x), label=\"second derivative\")\n", - "plt.plot(x, grad(grad(grad(f)))(x), label=\"third derivative\")\n", - "plt.legend()\n", - "plt.show()" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEDCAYAAAAhsS8XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nF3bGGIswXNgZYyzC/Be68EGj7hfMcwAAAABJRU5ErkJggg==\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - } - ] - }, - { - "metadata": { - "id": "-39gouo7mtgu", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Gradient tapes\n", - "\n", - "Every differentiable TensorFlow operation has an associated gradient function. For example, the gradient function of `tf.square(x)` would be a function that returns `2.0 * x`. To compute the gradient of a user-defined function (like `f(x)` in the example above), TensorFlow first \"records\" all the operations applied to compute the output of the function. We call this record a \"tape\". It then uses that tape and the gradients functions associated with each primitive operation to compute the gradients of the user-defined function using [reverse mode differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation).\n", - "\n", - "Since operations are recorded as they are executed, Python control flow (using `if`s and `while`s for example) is naturally handled:\n", - "\n" - ] - }, - { - "metadata": { - "id": "MH0UfjympWf7", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def f(x, y):\n", - " output = 1\n", - " for i in range(y):\n", - " output = tf.multiply(output, x)\n", - " return output\n", - "\n", - "def g(x, y):\n", - " # Return the gradient of `f` with respect to it's first parameter\n", - " return tfe.gradients_function(f)(x, y)[0]\n", - "\n", - "assert f(3.0, 2).numpy() == 9.0 # f(x, 2) is essentially x * x\n", - "assert g(3.0, 2).numpy() == 6.0 # And its gradient will be 2 * x\n", - "assert f(4.0, 3).numpy() == 64.0 # f(x, 3) is essentially x * x * x\n", - "assert g(4.0, 3).numpy() == 48.0 # And its gradient will be 3 * x * x" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "aNmR5-jhpX2t", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "At times it may be inconvenient to encapsulate computation of interest into a function. For example, if you want the gradient of the output with respect to intermediate values computed in the function. In such cases, the slightly more verbose but explicit [tf.GradientTape](https://www.tensorflow.org/api_docs/python/tf/GradientTape) context is useful. All computation inside the context of a `tf.GradientTape` is \"recorded\".\n", - "\n", - "For example:" - ] - }, - { - "metadata": { - "id": "bAFeIE8EuVIq", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "x = tf.ones((2, 2))\n", - " \n", - "# TODO(b/78880779): Remove the 'persistent=True' argument and use\n", - "# a single t.gradient() call when the bug is resolved.\n", - "with tf.GradientTape(persistent=True) as t:\n", - " # TODO(ashankar): Explain with \"watch\" argument better?\n", - " t.watch(x)\n", - " y = tf.reduce_sum(x)\n", - " z = tf.multiply(y, y)\n", - "\n", - "# Use the same tape to compute the derivative of z with respect to the\n", - "# intermediate value y.\n", - "dz_dy = t.gradient(z, y)\n", - "assert dz_dy.numpy() == 8.0\n", - "\n", - "# Derivative of z with respect to the original input tensor x\n", - "dz_dx = t.gradient(z, x)\n", - "for i in [0, 1]:\n", - " for j in [0, 1]:\n", - " assert dz_dx[i][j].numpy() == 8.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "DK05KXrAAld3", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Higher-order gradients\n", - "\n", - "Operations inside of the `GradientTape` context manager are recorded for automatic differentiation. If gradients are computed in that context, then the gradient computation is recorded as well. As a result, the exact same API works for higher-order gradients as well. For example:" - ] - }, - { - "metadata": { - "id": "cPQgthZ7ugRJ", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "# TODO(ashankar): Should we use the persistent tape here instead? Follow up on Tom and Alex's discussion\n", - "\n", - "x = tf.constant(1.0) # Convert the Python 1.0 to a Tensor object\n", - "\n", - "with tf.GradientTape() as t:\n", - " with tf.GradientTape() as t2:\n", - " t2.watch(x)\n", - " y = x * x * x\n", - " # Compute the gradient inside the 't' context manager\n", - " # which means the gradient computation is differentiable as well.\n", - " dy_dx = t2.gradient(y, x)\n", - "d2y_dx2 = t.gradient(dy_dx, x)\n", - "\n", - "assert dy_dx.numpy() == 3.0\n", - "assert d2y_dx2.numpy() == 6.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "4U1KKzUpNl58", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Next Steps\n", - "\n", - "In this tutorial we covered gradient computation in TensorFlow. With that we have enough of the primitives required to build an train neural networks, which we will cover in the [next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/3_neural_networks.ipynb)." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb deleted file mode 100644 index 37f2f4f402..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/3_datasets.ipynb +++ /dev/null @@ -1,261 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Datasets", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "collapsed_sections": [], - "toc_visible": true - } - }, - "cells": [ - { - "metadata": { - "id": "hvpDiXDuuNEu", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "4FWzUGc7uQKr", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "RRbeuncot80C", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Datasets" - ] - }, - { - "metadata": { - "id": "i9rtB6DnuFw5", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
\n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
" - ] - }, - { - "metadata": { - "id": "U9i2Dsh-ziXr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "This notebook demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your program. It covers:\n", - "\n", - "* Creating a `Dataset`.\n", - "* Iteration over a `Dataset` with eager execution enabled.\n", - "\n", - "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", - "\n", - "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", - "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", - "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." - ] - }, - { - "metadata": { - "id": "RlIWhyeLoYnG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "# Import TensorFlow.\n", - "import tensorflow as tf\n", - "\n", - "# Enable eager execution\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "H9UySOPLXdaw", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Create a source `Dataset`\n", - "\n", - "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." - ] - }, - { - "metadata": { - "id": "WPTUfGq6kJ5w", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", - "\n", - "# Create a CSV file\n", - "import tempfile\n", - "_, filename = tempfile.mkstemp()\n", - "with open(filename, 'w') as f:\n", - " f.write(\"\"\"Line 1\n", - "Line 2\n", - "Line 3\n", - " \"\"\")\n", - "ds_file = tf.data.TextLineDataset(filename)\n" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "twBfWd5xyu_d", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Apply transformations\n", - "\n", - "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." - ] - }, - { - "metadata": { - "id": "ngUe237Wt48W", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", - "ds_file = ds_file.batch(2)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "IDY4WsYRhP81", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Iterate\n", - "\n", - "When eager execution is enabled `Dataset` objects support iteration.\n", - "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." - ] - }, - { - "metadata": { - "id": "lCUWzso6mbqR", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "base_uri": "https://localhost:8080/", - "height": 153 - }, - "outputId": "8e4b0298-d27d-4ac7-e26a-ef94af0594ec", - "executionInfo": { - "elapsed": 388, - "status": "ok", - "timestamp": 1525154629129, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "print('Elements of ds_tensors:')\n", - "for x in ds_tensors:\n", - " print(x)\n", - "\n", - "print('\\nElements in ds_file:')\n", - "for x in ds_file:\n", - " print(x)" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Elements of ds_tensors:\n", - "tf.Tensor([1 9], shape=(2,), dtype=int32)\n", - "tf.Tensor([16 25], shape=(2,), dtype=int32)\n", - "tf.Tensor([ 4 36], shape=(2,), dtype=int32)\n", - "\n", - "Elements in ds_file:\n", - "tf.Tensor(['Line 1' 'Line 2'], shape=(2,), dtype=string)\n", - "tf.Tensor(['Line 3' ' '], shape=(2,), dtype=string)\n" - ], - "name": "stdout" - } - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb deleted file mode 100644 index ca0262eade..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/3_training_models.ipynb +++ /dev/null @@ -1,547 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Variables, models, and training", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "collapsed_sections": [], - "toc_visible": true - } - }, - "cells": [ - { - "metadata": { - "id": "5rmpybwysXGV", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "m8y3rGtQsYP2", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "hrXv0rU9sIma", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Variables, models, and training" - ] - }, - { - "metadata": { - "id": "7S0BwJ_8sLu7", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
\n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
" - ] - }, - { - "metadata": { - "id": "k2o3TTG4TFpt", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "In the previous tutorial we covered the TensorFlow APIs for automatic differentiation, a basic building block for machine learning.\n", - "In this tutorial we will use the TensorFlow primitives introduced in the prior tutorials to do some simple machine learning.\n", - "\n", - "TensorFlow also includes a higher-level neural networks API (`tf.keras`) which provides useful abstractions to reduce boilerplate. We strongly recommend those higher level APIs for people working with neural networks. However, in this short tutorial we cover neural network training from first principles to establish a strong foundation." - ] - }, - { - "metadata": { - "id": "3LXMVuV0VhDr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Setup" - ] - }, - { - "metadata": { - "id": "PJ64L90aVir3", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "tf.enable_eager_execution()\n", - "tfe = tf.contrib.eager # Shorthand for some symbols" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "eMAWbDJFVmMk", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Variables\n", - "\n", - "Tensors in TensorFlow are immutable stateless objects. Machine learning models, however, need to have changing state: as your model trains, the same code to compute predictions should behave differently over time (hopefully with a lower loss!). To represent this state which needs to change over the course of your computation, you can choose to rely on the fact that Python is a stateful programming language:\n" - ] - }, - { - "metadata": { - "id": "VkJwtLS_Jbn8", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "# Using python state\n", - "x = tf.zeros([10, 10])\n", - "x += 2 # This is equivalent to x = x + 2, which does not mutate the original\n", - " # value of x\n", - "print(x)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "wfneTXy7JcUz", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "TensorFlow, however, has stateful operations built in, and these are often more pleasant to use than low-level Python representations of your state. To represent weights in a model, for example, it's often convenient and efficient to use TensorFlow variables.\n", - "\n", - "A Variable is an object which stores a value and, when used in a TensorFlow computation, will implicitly read from this stored value. There are operations (`tf.assign_sub`, `tf.scatter_update`, etc) which manipulate the value stored in a TensorFlow variable." - ] - }, - { - "metadata": { - "id": "itxmrMil6DQi", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "v = tfe.Variable(1.0)\n", - "assert v.numpy() == 1.0\n", - "\n", - "# Re-assign the value\n", - "v.assign(3.0)\n", - "assert v.numpy() == 3.0\n", - "\n", - "# Use `v` in a TensorFlow operation like tf.square() and reassign\n", - "v.assign(tf.square(v))\n", - "assert v.numpy() == 9.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "-paSaeq1JzwC", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Computations using Variables are automatically traced when computing gradients. For Variables representing embeddings TensorFlow will do sparse updates by default, which are more computation and memory efficient.\n", - "\n", - "Using Variables is also a way to quickly let a reader of your code know that this piece of state is mutable." - ] - }, - { - "metadata": { - "id": "BMiFcDzE7Qu3", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Example: Fitting a linear model\n", - "\n", - "Let's now put the few concepts we have so far ---`Tensor`, `GradientTape`, `Variable` --- to build and train a simple model. This typically involves a few steps:\n", - "\n", - "1. Define the model.\n", - "2. Define a loss function.\n", - "3. Obtain training data.\n", - "4. Run through the training data and use an \"optimizer\" to adjust the variables to fit the data.\n", - "\n", - "In this tutorial, we'll walk through a trivial example of a simple linear model: `f(x) = x * W + b`, which has two variables - `W` and `b`. Furthermore, we'll synthesize data such that a well trained model would have `W = 3.0` and `b = 2.0`." - ] - }, - { - "metadata": { - "id": "gFzH64Jn9PIm", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define the model\n", - "\n", - "Let's define a simple class to encapsulate the variables and the computation." - ] - }, - { - "metadata": { - "id": "_WRu7Pze7wk8", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "class Model(object):\n", - " def __init__(self):\n", - " # Initialize variable to (5.0, 0.0)\n", - " # In practice, these should be initialized to random values.\n", - " self.W = tfe.Variable(5.0)\n", - " self.b = tfe.Variable(0.0)\n", - " \n", - " def __call__(self, x):\n", - " return self.W * x + self.b\n", - " \n", - "model = Model()\n", - "\n", - "assert model(3.0).numpy() == 15.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "xa6j_yXa-j79", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define a loss function\n", - "\n", - "A loss function measures how well the output of a model for a given input matches the desired output. Let's use the standard L2 loss." - ] - }, - { - "metadata": { - "id": "Y0ysUFGY924U", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def loss(predicted_y, desired_y):\n", - " return tf.reduce_mean(tf.square(predicted_y - desired_y))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "qutT_fkl_CBc", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Obtain training data\n", - "\n", - "Let's synthesize the training data with some noise." - ] - }, - { - "metadata": { - "id": "gxPTb-kt_N5m", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "TRUE_W = 3.0\n", - "TRUE_b = 2.0\n", - "NUM_EXAMPLES = 1000\n", - "\n", - "inputs = tf.random_normal(shape=[NUM_EXAMPLES])\n", - "noise = tf.random_normal(shape=[NUM_EXAMPLES])\n", - "outputs = inputs * TRUE_W + TRUE_b + noise" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "-50nq-wPBsAW", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Before we train the model let's visualize where the model stands right now. We'll plot the model's predictions in red and the training data in blue." - ] - }, - { - "metadata": { - "id": "_eb83LtrB4nt", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 293 - }, - "outputId": "3873f508-72fb-41e7-a7f5-3f513deefe38", - "executionInfo": { - "elapsed": 1210, - "status": "ok", - "timestamp": 1527005898290, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "import matplotlib.pyplot as plt\n", - "\n", - "plt.scatter(inputs, outputs, c='b')\n", - "plt.scatter(inputs, model(inputs), c='r')\n", - "plt.show()\n", - "\n", - "print('Current loss: '),\n", - "print(loss(model(inputs), outputs).numpy())" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEDCAYAAAA2k7/eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\noZkggi8IgtBMEMEXBEFoJvw//5K32R/vBHAAAAAASUVORK5CYII=\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - }, - { - "output_type": "stream", - "text": [ - "Current loss: 9.48636\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "sSDP-yeq_4jE", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define a training loop\n", - "\n", - "We now have our network and our training data. Let's train it, i.e., use the training data to update the model's variables (`W` and `b`) so that the loss goes down using [gradient descent](https://en.wikipedia.org/wiki/Gradient_descent). There are many variants of the gradient descent scheme that are captured in `tf.train.Optimizer` implementations. We'd highly recommend using those implementations, but in the spirit of building from first principles, in this particular example we will implement the basic math ourselves." - ] - }, - { - "metadata": { - "id": "MBIACgdnA55X", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def train(model, inputs, outputs, learning_rate):\n", - " with tf.GradientTape() as t:\n", - " current_loss = loss(model(inputs), outputs)\n", - " dW, db = t.gradient(current_loss, [model.W, model.b])\n", - " model.W.assign_sub(learning_rate * dW)\n", - " model.b.assign_sub(learning_rate * db)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "RwWPaJryD2aN", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Finally, let's repeatedly run through the training data and see how `W` and `b` evolve." - ] - }, - { - "metadata": { - "id": "XdfkR223D9dW", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 446 - }, - "outputId": "c43591ae-d5ac-4f2b-a8e7-bfce607e0919", - "executionInfo": { - "elapsed": 569, - "status": "ok", - "timestamp": 1527005915434, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "model = Model()\n", - "\n", - "# Collect the history of W-values and b-values to plot later\n", - "Ws, bs = [], []\n", - "epochs = range(10)\n", - "for epoch in epochs:\n", - " Ws.append(model.W.numpy())\n", - " bs.append(model.b.numpy())\n", - " current_loss = loss(model(inputs), outputs)\n", - "\n", - " train(model, inputs, outputs, learning_rate=0.1)\n", - " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", - " (epoch, Ws[-1], bs[-1], current_loss))\n", - "\n", - "# Let's plot it all\n", - "plt.plot(epochs, Ws, 'r',\n", - " epochs, bs, 'b')\n", - "plt.plot([TRUE_W] * len(epochs), 'r--',\n", - " [TRUE_b] * len(epochs), 'b--')\n", - "plt.legend(['W', 'b', 'true W', 'true_b'])\n", - "plt.show()\n", - " " - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Epoch 0: W=5.00 b=0.00, loss=9.48636\n", - "Epoch 1: W=4.58 b=0.42, loss=6.28101\n", - "Epoch 2: W=4.24 b=0.76, loss=4.29357\n", - "Epoch 3: W=3.98 b=1.02, loss=3.06128\n", - "Epoch 4: W=3.78 b=1.23, loss=2.29721\n", - "Epoch 5: W=3.61 b=1.39, loss=1.82345\n", - "Epoch 6: W=3.49 b=1.52, loss=1.52970\n", - "Epoch 7: W=3.38 b=1.62, loss=1.34756\n", - "Epoch 8: W=3.30 b=1.70, loss=1.23463\n", - "Epoch 9: W=3.24 b=1.76, loss=1.16460\n" - ], - "name": "stdout" - }, - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW0AAAEDCAYAAAD+/1UIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - } - ] - }, - { - "metadata": { - "id": "vPnIVuaSJwWz", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Next Steps\n", - "\n", - "In this tutorial we covered `Variable`s and built and trained a simple linear model using the TensorFlow primitives discussed so far.\n", - "\n", - "In theory, this is pretty much all you need to use TensorFlow for your machine learning research.\n", - "In practice, particularly for neural networks, the higher level APIs like `tf.keras` will be much more convenient since it provides higher level building blocks (called \"layers\"), utilities to save and restore state, a suite of loss functions, a suite of optimization strategies etc. \n", - "\n", - "The [next tutorial](TODO) will cover these higher level APIs." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb deleted file mode 100644 index 3378054be4..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/4_high_level.ipynb +++ /dev/null @@ -1,613 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Custom layers", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "collapsed_sections": [], - "toc_visible": true - }, - "kernelspec": { - "display_name": "Python 3", - "name": "python3" - } - }, - "cells": [ - { - "metadata": { - "id": "tDnwEv8FtJm7", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "JlknJBWQtKkI", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "60RdWsg1tETW", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Custom layers" - ] - }, - { - "metadata": { - "id": "BcJg7Enms86w", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
\n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
" - ] - }, - { - "metadata": { - "id": "UEu3q4jmpKVT", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "We recommend using `tf.keras` as a high-level API for building neural networks. That said, most TensorFlow APIs are usable with eager execution.\n" - ] - }, - { - "metadata": { - "id": "pwX7Fii1rwsJ", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "tfe = tf.contrib.eager\n", - "\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "zSFfVVjkrrsI", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Layers: common sets of useful operations\n", - "\n", - "Most of the time when writing code for machine learning models you want to operate at a higher level of abstraction than individual operations and manipulation of individual variables.\n", - "\n", - "Many machine learning models are expressible as the composition and stacking of relatively simple layers, and TensorFlow provides both a set of many common layers as a well as easy ways for you to write your own application-specific layers either from scratch or as the composition of existing layers.\n", - "\n", - "TensorFlow includes the full [Keras](https://keras.io) API in the tf.keras package, and the Keras layers are very useful when building your own models.\n" - ] - }, - { - "metadata": { - "id": "8PyXlPl-4TzQ", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "# In the tf.keras.layers package, layers are objects. To construct a layer,\n", - "# simply construct the object. Most layers take as a first argument the number\n", - "# of output dimensions / channels.\n", - "layer = tf.keras.layers.Dense(100)\n", - "# The number of input dimensions is often unnecessary, as it can be inferred\n", - "# the first time the layer is used, but it can be provided if you want to \n", - "# specify it manually, which is useful in some complex models.\n", - "layer = tf.keras.layers.Dense(10, input_shape=(None, 5))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "Fn69xxPO5Psr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "The full list of pre-existing layers can be seen in [the documentation](https://www.tensorflow.org/api_docs/python/tf/keras/layers). It includes Dense (a fully-connected layer),\n", - "Conv2D, LSTM, BatchNormalization, Dropout, and many others." - ] - }, - { - "metadata": { - "id": "E3XKNknP5Mhb", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 204 - }, - "outputId": "c5d52434-d980-4488-efa7-5660819d0207", - "executionInfo": { - "elapsed": 244, - "status": "ok", - "timestamp": 1527783641557, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "# To use a layer, simply call it.\n", - "layer(tf.zeros([10, 5]))" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 3 - } - ] - }, - { - "metadata": { - "id": "Wt_Nsv-L5t2s", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 221 - }, - "outputId": "f0d96dce-0128-4080-bfe2-0ee6fbc0ad90", - "executionInfo": { - "elapsed": 320, - "status": "ok", - "timestamp": 1527783642457, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "# Layers have many useful methods. For example, you can inspect all variables\n", - "# in a layer by calling layer.variables. In this case a fully-connected layer\n", - "# will have variables for weights and biases.\n", - "layer.variables" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "[,\n", - " ]" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 4 - } - ] - }, - { - "metadata": { - "id": "6ilvKjz8_4MQ", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 221 - }, - "outputId": "f647fced-c2d7-41a3-c237-242036784665", - "executionInfo": { - "elapsed": 226, - "status": "ok", - "timestamp": 1527783643252, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "# The variables are also accessible through nice accessors\n", - "layer.kernel, layer.bias" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "(,\n", - " )" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 5 - } - ] - }, - { - "metadata": { - "id": "O0kDbE54-5VS", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Implementing custom layers\n", - "The best way to implement your own layer is extending the tf.keras.Layer class and implementing:\n", - " * `__init__` , where you can do all input-independent initialization\n", - " * `build`, where you know the shapes of the input tensors and can do the rest of the initialization\n", - " * `call`, where you do the forward computation\n", - "\n", - "Note that you don't have to wait until `build` is called to create your variables, you can also create them in `__init__`. However, the advantage of creating them in `build` is that it enables late variable creation based on the shape of the inputs the layer will operate on. On the other hand, creating variables in `__init__` would mean that shapes required to create the variables will need to be explicitly specified." - ] - }, - { - "metadata": { - "id": "5Byl3n1k5kIy", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 391 - }, - "outputId": "6e7f9285-649a-4132-82ce-73ea92f15862", - "executionInfo": { - "elapsed": 251, - "status": "ok", - "timestamp": 1527783661512, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "class MyDenseLayer(tf.keras.layers.Layer):\n", - " def __init__(self, num_outputs):\n", - " super(MyDenseLayer, self).__init__()\n", - " self.num_outputs = num_outputs\n", - " \n", - " def build(self, input_shape):\n", - " self.kernel = self.add_variable(\"kernel\", \n", - " shape=[input_shape[-1].value, \n", - " self.num_outputs])\n", - " \n", - " def call(self, input):\n", - " return tf.matmul(input, self.kernel)\n", - " \n", - "layer = MyDenseLayer(10)\n", - "print(layer(tf.zeros([10, 5])))\n", - "print(layer.variables)" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(\n", - "[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(10, 10), dtype=float32)\n", - "[]\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "tk8E2vY0-z4Z", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Note that you don't have to wait until `build` is called to create your variables, you can also create them in `__init__`.\n", - "\n", - "Overall code is easier to read and maintain if it uses standard layers whenever possible, as other readers will be familiar with the behavior of standard layers. If you want to use a layer which is not present in tf.keras.layers or tf.contrib.layers, consider filing a [github issue](http://github.com/tensorflow/tensorflow/issues/new) or, even better, sending us a pull request!" - ] - }, - { - "metadata": { - "id": "Qhg4KlbKrs3G", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Models: composing layers\n", - "\n", - "Many interesting layer-like things in machine learning models are implemented by composing existing layers. For example, each residual block in a resnet is a composition of convolutions, batch normalizations, and a shortcut.\n", - "\n", - "The main class used when creating a layer-like thing which contains other layers is tf.keras.Model. Implementing one is done by inheriting from tf.keras.Model." - ] - }, - { - "metadata": { - "id": "N30DTXiRASlb", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "height": 190 - }, - "outputId": "a8b23a8e-5cf9-4bbf-f93b-6c763d74e2b3", - "executionInfo": { - "elapsed": 420, - "status": "ok", - "timestamp": 1527783698512, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - "class ResnetIdentityBlock(tf.keras.Model):\n", - " def __init__(self, kernel_size, filters):\n", - " super(ResnetIdentityBlock, self).__init__(name='')\n", - " filters1, filters2, filters3 = filters\n", - "\n", - " self.conv2a = tf.keras.layers.Conv2D(filters1, (1, 1))\n", - " self.bn2a = tf.keras.layers.BatchNormalization()\n", - "\n", - " self.conv2b = tf.keras.layers.Conv2D(filters2, kernel_size, padding='same')\n", - " self.bn2b = tf.keras.layers.BatchNormalization()\n", - "\n", - " self.conv2c = tf.keras.layers.Conv2D(filters3, (1, 1))\n", - " self.bn2c = tf.keras.layers.BatchNormalization()\n", - "\n", - " def call(self, input_tensor, training=False):\n", - " x = self.conv2a(input_tensor)\n", - " x = self.bn2a(x, training=training)\n", - " x = tf.nn.relu(x)\n", - "\n", - " x = self.conv2b(x)\n", - " x = self.bn2b(x, training=training)\n", - " x = tf.nn.relu(x)\n", - "\n", - " x = self.conv2c(x)\n", - " x = self.bn2c(x, training=training)\n", - "\n", - " x += input_tensor\n", - " return tf.nn.relu(x)\n", - "\n", - " \n", - "block = ResnetIdentityBlock(1, [1, 2, 3])\n", - "print(block(tf.zeros([1, 2, 3, 3])))\n", - "print([x.name for x in block.variables])" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(\n", - "[[[[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]\n", - "\n", - " [[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]]], shape=(1, 2, 3, 3), dtype=float32)\n", - "['resnet_identity_block_1/conv2d_3/kernel:0', 'resnet_identity_block_1/conv2d_3/bias:0', 'resnet_identity_block_1/batch_normalization_3/gamma:0', 'resnet_identity_block_1/batch_normalization_3/beta:0', 'resnet_identity_block_1/conv2d_4/kernel:0', 'resnet_identity_block_1/conv2d_4/bias:0', 'resnet_identity_block_1/batch_normalization_4/gamma:0', 'resnet_identity_block_1/batch_normalization_4/beta:0', 'resnet_identity_block_1/conv2d_5/kernel:0', 'resnet_identity_block_1/conv2d_5/bias:0', 'resnet_identity_block_1/batch_normalization_5/gamma:0', 'resnet_identity_block_1/batch_normalization_5/beta:0', 'resnet_identity_block_1/batch_normalization_3/moving_mean:0', 'resnet_identity_block_1/batch_normalization_3/moving_variance:0', 'resnet_identity_block_1/batch_normalization_4/moving_mean:0', 'resnet_identity_block_1/batch_normalization_4/moving_variance:0', 'resnet_identity_block_1/batch_normalization_5/moving_mean:0', 'resnet_identity_block_1/batch_normalization_5/moving_variance:0']\n" - ], - "name": "stdout" - } - ] - }, - { - "metadata": { - "id": "wYfucVw65PMj", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Much of the time, however, models which compose many layers simply call one layer after the other. This can be done in very little code using tf.keras.Sequential" - ] - }, - { - "metadata": { - "id": "L9frk7Ur4uvJ", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - }, - "base_uri": "https://localhost:8080/", - "height": 153 - }, - "outputId": "882e9076-b6d9-4380-bb1e-7c6b57d54c39", - "executionInfo": { - "elapsed": 361, - "status": "ok", - "timestamp": 1526674830777, - "user": { - "displayName": "Alexandre Passos", - "photoUrl": "//lh4.googleusercontent.com/-kmTTWXEgAPw/AAAAAAAAAAI/AAAAAAAAAC0/q_DoOzKGwds/s50-c-k-no/photo.jpg", - "userId": "108023195365833072773" - }, - "user_tz": 420 - } - }, - "cell_type": "code", - "source": [ - " my_seq = tf.keras.Sequential([tf.keras.layers.Conv2D(1, (1, 1)),\n", - " tf.keras.layers.BatchNormalization(),\n", - " tf.keras.layers.Conv2D(2, 1, \n", - " padding='same'),\n", - " tf.keras.layers.BatchNormalization(),\n", - " tf.keras.layers.Conv2D(3, (1, 1)),\n", - " tf.keras.layers.BatchNormalization()])\n", - "my_seq(tf.zeros([1, 2, 3, 3]))" - ], - "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 26 - } - ] - }, - { - "metadata": { - "id": "c5YwYcnuK-wc", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Next steps\n", - "\n", - "Now you can go back to the previous notebook and adapt the linear regression example to use layers and models to be better structured." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/README.md b/tensorflow/contrib/eager/python/examples/notebooks/README.md new file mode 100644 index 0000000000..2a833bc683 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/README.md @@ -0,0 +1,11 @@ +## Research and experimentation + +Eager execution provides an imperative, define-by-run interface for advanced +operations. Write custom layers, forward passes, and training loops with auto +differentiation. Start with these notebooks, then read the +[eager execution guide](https://www.tensorflow.org/guide/eager). + +1. [Eager execution basics](./eager_basics.ipynb) +2. [Automatic differentiation and gradient tapes](./automatic_differentiation.ipynb) +3. [Variables, models, and training](./variables_models_training.ipynb) +4. [Custom layers](./custom_layers.ipynb) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb new file mode 100644 index 0000000000..d7bcf749e8 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb @@ -0,0 +1,385 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Automatic differentiation and gradient tape", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, + "cells": [ + { + "metadata": { + "id": "t09eeeR5prIJ", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "GCCk8_dHpuNf", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "xh8WkEwWpnm7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Automatic differentiation and gradient tape" + ] + }, + { + "metadata": { + "id": "idv0bPeCp325", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "vDJ4XzMqodTy", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "In the previous tutorial we introduced `Tensor`s and operations on them. In this tutorial we will cover [automatic differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation), a key technique for optimizing machine learning models." + ] + }, + { + "metadata": { + "id": "GQJysDM__Qb0", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Setup\n" + ] + }, + { + "metadata": { + "id": "OiMPZStlibBv", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "tf.enable_eager_execution()\n", + "\n", + "tfe = tf.contrib.eager # Shorthand for some symbols" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "1CLWJl0QliB0", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Derivatives of a function\n", + "\n", + "TensorFlow provides APIs for automatic differentiation - computing the derivative of a function. The way that more closely mimics the math is to encapsulate the computation in a Python function, say `f`, and use `tfe.gradients_function` to create a function that computes the derivatives of `f` with respect to its arguments. If you're familiar with [autograd](https://github.com/HIPS/autograd) for differentiating numpy functions, this will be familiar. For example: " + ] + }, + { + "metadata": { + "id": "9FViq92UX7P8", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "from math import pi\n", + "\n", + "def f(x):\n", + " return tf.square(tf.sin(x))\n", + "\n", + "assert f(pi/2).numpy() == 1.0\n", + "\n", + "\n", + "# grad_f will return a list of derivatives of f\n", + "# with respect to its arguments. Since f() has a single argument,\n", + "# grad_f will return a list with a single element.\n", + "grad_f = tfe.gradients_function(f)\n", + "assert tf.abs(grad_f(pi/2)[0]).numpy() < 1e-7" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "v9fPs8RyopCf", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Higher-order gradients\n", + "\n", + "The same API can be used to differentiate as many times as you like:\n" + ] + }, + { + "metadata": { + "id": "3D0ZvnGYo0rW", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 276 + }, + "outputId": "e23f8cc6-6813-4944-f20f-825b8a03c2ff", + "executionInfo": { + "elapsed": 730, + "status": "ok", + "timestamp": 1527005655565, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "def f(x):\n", + " return tf.square(tf.sin(x))\n", + "\n", + "def grad(f):\n", + " return lambda x: tfe.gradients_function(f)(x)[0]\n", + "\n", + "x = tf.lin_space(-2*pi, 2*pi, 100) # 100 points between -2Ï€ and +2Ï€\n", + "\n", + "import matplotlib.pyplot as plt\n", + "\n", + "plt.plot(x, f(x), label=\"f\")\n", + "plt.plot(x, grad(f)(x), label=\"first derivative\")\n", + "plt.plot(x, grad(grad(f))(x), label=\"second derivative\")\n", + "plt.plot(x, grad(grad(grad(f)))(x), label=\"third derivative\")\n", + "plt.legend()\n", + "plt.show()" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "display_data", + "data": { + "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEDCAYAAAAhsS8XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nF3bGGIswXNgZYyzC/Be68EGj7hfMcwAAAABJRU5ErkJggg==\n", + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + } + } + ] + }, + { + "metadata": { + "id": "-39gouo7mtgu", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Gradient tapes\n", + "\n", + "Every differentiable TensorFlow operation has an associated gradient function. For example, the gradient function of `tf.square(x)` would be a function that returns `2.0 * x`. To compute the gradient of a user-defined function (like `f(x)` in the example above), TensorFlow first \"records\" all the operations applied to compute the output of the function. We call this record a \"tape\". It then uses that tape and the gradients functions associated with each primitive operation to compute the gradients of the user-defined function using [reverse mode differentiation](https://en.wikipedia.org/wiki/Automatic_differentiation).\n", + "\n", + "Since operations are recorded as they are executed, Python control flow (using `if`s and `while`s for example) is naturally handled:\n", + "\n" + ] + }, + { + "metadata": { + "id": "MH0UfjympWf7", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def f(x, y):\n", + " output = 1\n", + " for i in range(y):\n", + " output = tf.multiply(output, x)\n", + " return output\n", + "\n", + "def g(x, y):\n", + " # Return the gradient of `f` with respect to it's first parameter\n", + " return tfe.gradients_function(f)(x, y)[0]\n", + "\n", + "assert f(3.0, 2).numpy() == 9.0 # f(x, 2) is essentially x * x\n", + "assert g(3.0, 2).numpy() == 6.0 # And its gradient will be 2 * x\n", + "assert f(4.0, 3).numpy() == 64.0 # f(x, 3) is essentially x * x * x\n", + "assert g(4.0, 3).numpy() == 48.0 # And its gradient will be 3 * x * x" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "aNmR5-jhpX2t", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "At times it may be inconvenient to encapsulate computation of interest into a function. For example, if you want the gradient of the output with respect to intermediate values computed in the function. In such cases, the slightly more verbose but explicit [tf.GradientTape](https://www.tensorflow.org/api_docs/python/tf/GradientTape) context is useful. All computation inside the context of a `tf.GradientTape` is \"recorded\".\n", + "\n", + "For example:" + ] + }, + { + "metadata": { + "id": "bAFeIE8EuVIq", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "x = tf.ones((2, 2))\n", + " \n", + "# TODO(b/78880779): Remove the 'persistent=True' argument and use\n", + "# a single t.gradient() call when the bug is resolved.\n", + "with tf.GradientTape(persistent=True) as t:\n", + " # TODO(ashankar): Explain with \"watch\" argument better?\n", + " t.watch(x)\n", + " y = tf.reduce_sum(x)\n", + " z = tf.multiply(y, y)\n", + "\n", + "# Use the same tape to compute the derivative of z with respect to the\n", + "# intermediate value y.\n", + "dz_dy = t.gradient(z, y)\n", + "assert dz_dy.numpy() == 8.0\n", + "\n", + "# Derivative of z with respect to the original input tensor x\n", + "dz_dx = t.gradient(z, x)\n", + "for i in [0, 1]:\n", + " for j in [0, 1]:\n", + " assert dz_dx[i][j].numpy() == 8.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "DK05KXrAAld3", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Higher-order gradients\n", + "\n", + "Operations inside of the `GradientTape` context manager are recorded for automatic differentiation. If gradients are computed in that context, then the gradient computation is recorded as well. As a result, the exact same API works for higher-order gradients as well. For example:" + ] + }, + { + "metadata": { + "id": "cPQgthZ7ugRJ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "# TODO(ashankar): Should we use the persistent tape here instead? Follow up on Tom and Alex's discussion\n", + "\n", + "x = tf.constant(1.0) # Convert the Python 1.0 to a Tensor object\n", + "\n", + "with tf.GradientTape() as t:\n", + " with tf.GradientTape() as t2:\n", + " t2.watch(x)\n", + " y = x * x * x\n", + " # Compute the gradient inside the 't' context manager\n", + " # which means the gradient computation is differentiable as well.\n", + " dy_dx = t2.gradient(y, x)\n", + "d2y_dx2 = t.gradient(dy_dx, x)\n", + "\n", + "assert dy_dx.numpy() == 3.0\n", + "assert d2y_dx2.numpy() == 6.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "4U1KKzUpNl58", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Next Steps\n", + "\n", + "In this tutorial we covered gradient computation in TensorFlow. With that we have enough of the primitives required to build an train neural networks, which we will cover in the [next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/3_neural_networks.ipynb)." + ] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb new file mode 100644 index 0000000000..3378054be4 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb @@ -0,0 +1,613 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Custom layers", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + }, + "kernelspec": { + "display_name": "Python 3", + "name": "python3" + } + }, + "cells": [ + { + "metadata": { + "id": "tDnwEv8FtJm7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "JlknJBWQtKkI", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "60RdWsg1tETW", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Custom layers" + ] + }, + { + "metadata": { + "id": "BcJg7Enms86w", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "UEu3q4jmpKVT", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "We recommend using `tf.keras` as a high-level API for building neural networks. That said, most TensorFlow APIs are usable with eager execution.\n" + ] + }, + { + "metadata": { + "id": "pwX7Fii1rwsJ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "tfe = tf.contrib.eager\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "zSFfVVjkrrsI", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Layers: common sets of useful operations\n", + "\n", + "Most of the time when writing code for machine learning models you want to operate at a higher level of abstraction than individual operations and manipulation of individual variables.\n", + "\n", + "Many machine learning models are expressible as the composition and stacking of relatively simple layers, and TensorFlow provides both a set of many common layers as a well as easy ways for you to write your own application-specific layers either from scratch or as the composition of existing layers.\n", + "\n", + "TensorFlow includes the full [Keras](https://keras.io) API in the tf.keras package, and the Keras layers are very useful when building your own models.\n" + ] + }, + { + "metadata": { + "id": "8PyXlPl-4TzQ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "# In the tf.keras.layers package, layers are objects. To construct a layer,\n", + "# simply construct the object. Most layers take as a first argument the number\n", + "# of output dimensions / channels.\n", + "layer = tf.keras.layers.Dense(100)\n", + "# The number of input dimensions is often unnecessary, as it can be inferred\n", + "# the first time the layer is used, but it can be provided if you want to \n", + "# specify it manually, which is useful in some complex models.\n", + "layer = tf.keras.layers.Dense(10, input_shape=(None, 5))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "Fn69xxPO5Psr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "The full list of pre-existing layers can be seen in [the documentation](https://www.tensorflow.org/api_docs/python/tf/keras/layers). It includes Dense (a fully-connected layer),\n", + "Conv2D, LSTM, BatchNormalization, Dropout, and many others." + ] + }, + { + "metadata": { + "id": "E3XKNknP5Mhb", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 204 + }, + "outputId": "c5d52434-d980-4488-efa7-5660819d0207", + "executionInfo": { + "elapsed": 244, + "status": "ok", + "timestamp": 1527783641557, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "# To use a layer, simply call it.\n", + "layer(tf.zeros([10, 5]))" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 3 + } + ] + }, + { + "metadata": { + "id": "Wt_Nsv-L5t2s", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 221 + }, + "outputId": "f0d96dce-0128-4080-bfe2-0ee6fbc0ad90", + "executionInfo": { + "elapsed": 320, + "status": "ok", + "timestamp": 1527783642457, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "# Layers have many useful methods. For example, you can inspect all variables\n", + "# in a layer by calling layer.variables. In this case a fully-connected layer\n", + "# will have variables for weights and biases.\n", + "layer.variables" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "[,\n", + " ]" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 4 + } + ] + }, + { + "metadata": { + "id": "6ilvKjz8_4MQ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 221 + }, + "outputId": "f647fced-c2d7-41a3-c237-242036784665", + "executionInfo": { + "elapsed": 226, + "status": "ok", + "timestamp": 1527783643252, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "# The variables are also accessible through nice accessors\n", + "layer.kernel, layer.bias" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "(,\n", + " )" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 5 + } + ] + }, + { + "metadata": { + "id": "O0kDbE54-5VS", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Implementing custom layers\n", + "The best way to implement your own layer is extending the tf.keras.Layer class and implementing:\n", + " * `__init__` , where you can do all input-independent initialization\n", + " * `build`, where you know the shapes of the input tensors and can do the rest of the initialization\n", + " * `call`, where you do the forward computation\n", + "\n", + "Note that you don't have to wait until `build` is called to create your variables, you can also create them in `__init__`. However, the advantage of creating them in `build` is that it enables late variable creation based on the shape of the inputs the layer will operate on. On the other hand, creating variables in `__init__` would mean that shapes required to create the variables will need to be explicitly specified." + ] + }, + { + "metadata": { + "id": "5Byl3n1k5kIy", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 391 + }, + "outputId": "6e7f9285-649a-4132-82ce-73ea92f15862", + "executionInfo": { + "elapsed": 251, + "status": "ok", + "timestamp": 1527783661512, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "class MyDenseLayer(tf.keras.layers.Layer):\n", + " def __init__(self, num_outputs):\n", + " super(MyDenseLayer, self).__init__()\n", + " self.num_outputs = num_outputs\n", + " \n", + " def build(self, input_shape):\n", + " self.kernel = self.add_variable(\"kernel\", \n", + " shape=[input_shape[-1].value, \n", + " self.num_outputs])\n", + " \n", + " def call(self, input):\n", + " return tf.matmul(input, self.kernel)\n", + " \n", + "layer = MyDenseLayer(10)\n", + "print(layer(tf.zeros([10, 5])))\n", + "print(layer.variables)" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "tf.Tensor(\n", + "[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", + " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(10, 10), dtype=float32)\n", + "[]\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "tk8E2vY0-z4Z", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Note that you don't have to wait until `build` is called to create your variables, you can also create them in `__init__`.\n", + "\n", + "Overall code is easier to read and maintain if it uses standard layers whenever possible, as other readers will be familiar with the behavior of standard layers. If you want to use a layer which is not present in tf.keras.layers or tf.contrib.layers, consider filing a [github issue](http://github.com/tensorflow/tensorflow/issues/new) or, even better, sending us a pull request!" + ] + }, + { + "metadata": { + "id": "Qhg4KlbKrs3G", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Models: composing layers\n", + "\n", + "Many interesting layer-like things in machine learning models are implemented by composing existing layers. For example, each residual block in a resnet is a composition of convolutions, batch normalizations, and a shortcut.\n", + "\n", + "The main class used when creating a layer-like thing which contains other layers is tf.keras.Model. Implementing one is done by inheriting from tf.keras.Model." + ] + }, + { + "metadata": { + "id": "N30DTXiRASlb", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 190 + }, + "outputId": "a8b23a8e-5cf9-4bbf-f93b-6c763d74e2b3", + "executionInfo": { + "elapsed": 420, + "status": "ok", + "timestamp": 1527783698512, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "class ResnetIdentityBlock(tf.keras.Model):\n", + " def __init__(self, kernel_size, filters):\n", + " super(ResnetIdentityBlock, self).__init__(name='')\n", + " filters1, filters2, filters3 = filters\n", + "\n", + " self.conv2a = tf.keras.layers.Conv2D(filters1, (1, 1))\n", + " self.bn2a = tf.keras.layers.BatchNormalization()\n", + "\n", + " self.conv2b = tf.keras.layers.Conv2D(filters2, kernel_size, padding='same')\n", + " self.bn2b = tf.keras.layers.BatchNormalization()\n", + "\n", + " self.conv2c = tf.keras.layers.Conv2D(filters3, (1, 1))\n", + " self.bn2c = tf.keras.layers.BatchNormalization()\n", + "\n", + " def call(self, input_tensor, training=False):\n", + " x = self.conv2a(input_tensor)\n", + " x = self.bn2a(x, training=training)\n", + " x = tf.nn.relu(x)\n", + "\n", + " x = self.conv2b(x)\n", + " x = self.bn2b(x, training=training)\n", + " x = tf.nn.relu(x)\n", + "\n", + " x = self.conv2c(x)\n", + " x = self.bn2c(x, training=training)\n", + "\n", + " x += input_tensor\n", + " return tf.nn.relu(x)\n", + "\n", + " \n", + "block = ResnetIdentityBlock(1, [1, 2, 3])\n", + "print(block(tf.zeros([1, 2, 3, 3])))\n", + "print([x.name for x in block.variables])" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "tf.Tensor(\n", + "[[[[ 0. 0. 0.]\n", + " [ 0. 0. 0.]\n", + " [ 0. 0. 0.]]\n", + "\n", + " [[ 0. 0. 0.]\n", + " [ 0. 0. 0.]\n", + " [ 0. 0. 0.]]]], shape=(1, 2, 3, 3), dtype=float32)\n", + "['resnet_identity_block_1/conv2d_3/kernel:0', 'resnet_identity_block_1/conv2d_3/bias:0', 'resnet_identity_block_1/batch_normalization_3/gamma:0', 'resnet_identity_block_1/batch_normalization_3/beta:0', 'resnet_identity_block_1/conv2d_4/kernel:0', 'resnet_identity_block_1/conv2d_4/bias:0', 'resnet_identity_block_1/batch_normalization_4/gamma:0', 'resnet_identity_block_1/batch_normalization_4/beta:0', 'resnet_identity_block_1/conv2d_5/kernel:0', 'resnet_identity_block_1/conv2d_5/bias:0', 'resnet_identity_block_1/batch_normalization_5/gamma:0', 'resnet_identity_block_1/batch_normalization_5/beta:0', 'resnet_identity_block_1/batch_normalization_3/moving_mean:0', 'resnet_identity_block_1/batch_normalization_3/moving_variance:0', 'resnet_identity_block_1/batch_normalization_4/moving_mean:0', 'resnet_identity_block_1/batch_normalization_4/moving_variance:0', 'resnet_identity_block_1/batch_normalization_5/moving_mean:0', 'resnet_identity_block_1/batch_normalization_5/moving_variance:0']\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "wYfucVw65PMj", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Much of the time, however, models which compose many layers simply call one layer after the other. This can be done in very little code using tf.keras.Sequential" + ] + }, + { + "metadata": { + "id": "L9frk7Ur4uvJ", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "base_uri": "https://localhost:8080/", + "height": 153 + }, + "outputId": "882e9076-b6d9-4380-bb1e-7c6b57d54c39", + "executionInfo": { + "elapsed": 361, + "status": "ok", + "timestamp": 1526674830777, + "user": { + "displayName": "Alexandre Passos", + "photoUrl": "//lh4.googleusercontent.com/-kmTTWXEgAPw/AAAAAAAAAAI/AAAAAAAAAC0/q_DoOzKGwds/s50-c-k-no/photo.jpg", + "userId": "108023195365833072773" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + " my_seq = tf.keras.Sequential([tf.keras.layers.Conv2D(1, (1, 1)),\n", + " tf.keras.layers.BatchNormalization(),\n", + " tf.keras.layers.Conv2D(2, 1, \n", + " padding='same'),\n", + " tf.keras.layers.BatchNormalization(),\n", + " tf.keras.layers.Conv2D(3, (1, 1)),\n", + " tf.keras.layers.BatchNormalization()])\n", + "my_seq(tf.zeros([1, 2, 3, 3]))" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 26 + } + ] + }, + { + "metadata": { + "id": "c5YwYcnuK-wc", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Next steps\n", + "\n", + "Now you can go back to the previous notebook and adapt the linear regression example to use layers and models to be better structured." + ] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb new file mode 100644 index 0000000000..37f2f4f402 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb @@ -0,0 +1,261 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Datasets", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, + "cells": [ + { + "metadata": { + "id": "hvpDiXDuuNEu", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "4FWzUGc7uQKr", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "RRbeuncot80C", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Datasets" + ] + }, + { + "metadata": { + "id": "i9rtB6DnuFw5", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "This notebook demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your program. It covers:\n", + "\n", + "* Creating a `Dataset`.\n", + "* Iteration over a `Dataset` with eager execution enabled.\n", + "\n", + "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", + "\n", + "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", + "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", + "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." + ] + }, + { + "metadata": { + "id": "RlIWhyeLoYnG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "# Import TensorFlow.\n", + "import tensorflow as tf\n", + "\n", + "# Enable eager execution\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "H9UySOPLXdaw", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Create a source `Dataset`\n", + "\n", + "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." + ] + }, + { + "metadata": { + "id": "WPTUfGq6kJ5w", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", + "\n", + "# Create a CSV file\n", + "import tempfile\n", + "_, filename = tempfile.mkstemp()\n", + "with open(filename, 'w') as f:\n", + " f.write(\"\"\"Line 1\n", + "Line 2\n", + "Line 3\n", + " \"\"\")\n", + "ds_file = tf.data.TextLineDataset(filename)\n" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "twBfWd5xyu_d", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Apply transformations\n", + "\n", + "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." + ] + }, + { + "metadata": { + "id": "ngUe237Wt48W", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", + "ds_file = ds_file.batch(2)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "IDY4WsYRhP81", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Iterate\n", + "\n", + "When eager execution is enabled `Dataset` objects support iteration.\n", + "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." + ] + }, + { + "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "base_uri": "https://localhost:8080/", + "height": 153 + }, + "outputId": "8e4b0298-d27d-4ac7-e26a-ef94af0594ec", + "executionInfo": { + "elapsed": 388, + "status": "ok", + "timestamp": 1525154629129, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "print('Elements of ds_tensors:')\n", + "for x in ds_tensors:\n", + " print(x)\n", + "\n", + "print('\\nElements in ds_file:')\n", + "for x in ds_file:\n", + " print(x)" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "Elements of ds_tensors:\n", + "tf.Tensor([1 9], shape=(2,), dtype=int32)\n", + "tf.Tensor([16 25], shape=(2,), dtype=int32)\n", + "tf.Tensor([ 4 36], shape=(2,), dtype=int32)\n", + "\n", + "Elements in ds_file:\n", + "tf.Tensor(['Line 1' 'Line 2'], shape=(2,), dtype=string)\n", + "tf.Tensor(['Line 3' ' '], shape=(2,), dtype=string)\n" + ], + "name": "stdout" + } + ] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb new file mode 100644 index 0000000000..99d141bf2b --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb @@ -0,0 +1,491 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Eager execution basics", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, + "cells": [ + { + "metadata": { + "id": "iPpI7RaYoZuE", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "hro2InpHobKk", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Eager execution basics" + ] + }, + { + "metadata": { + "id": "Hndw-YcxoOJK", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "6sILUVbHoSgH", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "This is an introductory tutorial for using TensorFlow. It will cover:\n", + "\n", + "* Importing required packages\n", + "* Creating and using Tensors\n", + "* Using GPU acceleration" + ] + }, + { + "metadata": { + "id": "z1JcS5iBXMRO", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Import TensorFlow\n", + "\n", + "To get started, import the `tensorflow` module and enable eager execution.\n", + "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." + ] + }, + { + "metadata": { + "id": "RlIWhyeLoYnG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "H9UySOPLXdaw", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Tensors\n", + "\n", + "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" + ] + }, + { + "metadata": { + "id": "ngUe237Wt48W", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 125 + }, + "cellView": "code", + "outputId": "b1a1cd60-4eb3-443d-cd6b-68406390784e", + "executionInfo": { + "elapsed": 320, + "status": "ok", + "timestamp": 1526420535530, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "print(tf.add(1, 2))\n", + "print(tf.add([1, 2], [3, 4]))\n", + "print(tf.square(5))\n", + "print(tf.reduce_sum([1, 2, 3]))\n", + "print(tf.encode_base64(\"hello world\"))\n", + "\n", + "# Operator overloading is also supported\n", + "print(tf.square(2) + tf.square(3))" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "tf.Tensor(3, shape=(), dtype=int32)\n", + "tf.Tensor([4 6], shape=(2,), dtype=int32)\n", + "tf.Tensor(25, shape=(), dtype=int32)\n", + "tf.Tensor(6, shape=(), dtype=int32)\n", + "tf.Tensor(aGVsbG8gd29ybGQ, shape=(), dtype=string)\n", + "tf.Tensor(13, shape=(), dtype=int32)\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "IDY4WsYRhP81", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Each Tensor has a shape and a datatype" + ] + }, + { + "metadata": { + "id": "srYWH1MdJNG7", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 53 + }, + "outputId": "5e4ac41c-5115-4e50-eba0-42e249c16561", + "executionInfo": { + "elapsed": 215, + "status": "ok", + "timestamp": 1526420538162, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "x = tf.matmul([[1]], [[2, 3]])\n", + "print(x.shape)\n", + "print(x.dtype)" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "(1, 2)\n", + "\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "eBPw8e8vrsom", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", + "\n", + "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", + "2. Tensors are immutable." + ] + }, + { + "metadata": { + "id": "Dwi1tdW3JBw6", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### NumPy Compatibility\n", + "\n", + "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", + "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", + "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", + "\n", + "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", + "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." + ] + }, + { + "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 251 + }, + "outputId": "fd0a22bc-8249-49dd-fcbd-63161cc47e46", + "executionInfo": { + "elapsed": 238, + "status": "ok", + "timestamp": 1526420540562, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "import numpy as np\n", + "\n", + "ndarray = np.ones([3, 3])\n", + "\n", + "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", + "tensor = tf.multiply(ndarray, 42)\n", + "print(tensor)\n", + "\n", + "\n", + "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", + "print(np.add(tensor, 1))\n", + "\n", + "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", + "print(tensor.numpy())" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "TensorFlow operations convert numpy arrays to Tensors automatically\n", + "tf.Tensor(\n", + "[[ 42. 42. 42.]\n", + " [ 42. 42. 42.]\n", + " [ 42. 42. 42.]], shape=(3, 3), dtype=float64)\n", + "And NumPy operations convert Tensors to numpy arrays automatically\n", + "[[ 43. 43. 43.]\n", + " [ 43. 43. 43.]\n", + " [ 43. 43. 43.]]\n", + "The .numpy() method explicitly converts a Tensor to a numpy array\n", + "[[ 42. 42. 42.]\n", + " [ 42. 42. 42.]\n", + " [ 42. 42. 42.]]\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "PBNP8yTRfu_X", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## GPU acceleration\n", + "\n", + "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" + ] + }, + { + "metadata": { + "id": "3Twf_Rw-gQFM", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 53 + }, + "cellView": "code", + "outputId": "2239ae2b-adf3-4895-b1f3-464cf5361d1b", + "executionInfo": { + "elapsed": 340, + "status": "ok", + "timestamp": 1526420543562, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "x = tf.random_uniform([3, 3])\n", + "\n", + "print(\"Is there a GPU available: \"),\n", + "print(tf.test.is_gpu_available())\n", + "\n", + "print(\"Is the Tensor on GPU #0: \"),\n", + "print(x.device.endswith('GPU:0'))" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "Is there a GPU available: False\n", + "Is the Tensor on GPU #0: False\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "vpgYzgVXW2Ud", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Device Names\n", + "\n", + "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." + ] + }, + { + "metadata": { + "id": "ZWZQCimzuqyP", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "\n", + "\n", + "### Explicit Device Placement\n", + "\n", + "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" + ] + }, + { + "metadata": { + "id": "RjkNZTuauy-Q", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 53 + }, + "outputId": "2e613293-ccac-4db2-b793-8ceb5b5adcfd", + "executionInfo": { + "elapsed": 1762, + "status": "ok", + "timestamp": 1526420547562, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "def time_matmul(x):\n", + " %timeit tf.matmul(x, x)\n", + "\n", + "# Force execution on CPU\n", + "print(\"On CPU:\")\n", + "with tf.device(\"CPU:0\"):\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"CPU:0\")\n", + " time_matmul(x)\n", + "\n", + "# Force execution on GPU #0 if available\n", + "if tf.test.is_gpu_available():\n", + " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"GPU:0\")\n", + " time_matmul(x)" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "On CPU:\n", + "10 loops, best of 3: 35.8 ms per loop\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "YEOJTNiOvnpQ", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Next Steps\n", + "\n", + "In this tutorial we covered the most fundamental concepts in TensorFlow - `Tensor`s, operations, and devices.\n", + "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." + ] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb new file mode 100644 index 0000000000..ca0262eade --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb @@ -0,0 +1,547 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Variables, models, and training", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "collapsed_sections": [], + "toc_visible": true + } + }, + "cells": [ + { + "metadata": { + "id": "5rmpybwysXGV", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "m8y3rGtQsYP2", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "hrXv0rU9sIma", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Variables, models, and training" + ] + }, + { + "metadata": { + "id": "7S0BwJ_8sLu7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
\n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
" + ] + }, + { + "metadata": { + "id": "k2o3TTG4TFpt", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "In the previous tutorial we covered the TensorFlow APIs for automatic differentiation, a basic building block for machine learning.\n", + "In this tutorial we will use the TensorFlow primitives introduced in the prior tutorials to do some simple machine learning.\n", + "\n", + "TensorFlow also includes a higher-level neural networks API (`tf.keras`) which provides useful abstractions to reduce boilerplate. We strongly recommend those higher level APIs for people working with neural networks. However, in this short tutorial we cover neural network training from first principles to establish a strong foundation." + ] + }, + { + "metadata": { + "id": "3LXMVuV0VhDr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Setup" + ] + }, + { + "metadata": { + "id": "PJ64L90aVir3", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "tf.enable_eager_execution()\n", + "tfe = tf.contrib.eager # Shorthand for some symbols" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "eMAWbDJFVmMk", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Variables\n", + "\n", + "Tensors in TensorFlow are immutable stateless objects. Machine learning models, however, need to have changing state: as your model trains, the same code to compute predictions should behave differently over time (hopefully with a lower loss!). To represent this state which needs to change over the course of your computation, you can choose to rely on the fact that Python is a stateful programming language:\n" + ] + }, + { + "metadata": { + "id": "VkJwtLS_Jbn8", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "# Using python state\n", + "x = tf.zeros([10, 10])\n", + "x += 2 # This is equivalent to x = x + 2, which does not mutate the original\n", + " # value of x\n", + "print(x)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "wfneTXy7JcUz", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "TensorFlow, however, has stateful operations built in, and these are often more pleasant to use than low-level Python representations of your state. To represent weights in a model, for example, it's often convenient and efficient to use TensorFlow variables.\n", + "\n", + "A Variable is an object which stores a value and, when used in a TensorFlow computation, will implicitly read from this stored value. There are operations (`tf.assign_sub`, `tf.scatter_update`, etc) which manipulate the value stored in a TensorFlow variable." + ] + }, + { + "metadata": { + "id": "itxmrMil6DQi", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "v = tfe.Variable(1.0)\n", + "assert v.numpy() == 1.0\n", + "\n", + "# Re-assign the value\n", + "v.assign(3.0)\n", + "assert v.numpy() == 3.0\n", + "\n", + "# Use `v` in a TensorFlow operation like tf.square() and reassign\n", + "v.assign(tf.square(v))\n", + "assert v.numpy() == 9.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "-paSaeq1JzwC", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Computations using Variables are automatically traced when computing gradients. For Variables representing embeddings TensorFlow will do sparse updates by default, which are more computation and memory efficient.\n", + "\n", + "Using Variables is also a way to quickly let a reader of your code know that this piece of state is mutable." + ] + }, + { + "metadata": { + "id": "BMiFcDzE7Qu3", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Example: Fitting a linear model\n", + "\n", + "Let's now put the few concepts we have so far ---`Tensor`, `GradientTape`, `Variable` --- to build and train a simple model. This typically involves a few steps:\n", + "\n", + "1. Define the model.\n", + "2. Define a loss function.\n", + "3. Obtain training data.\n", + "4. Run through the training data and use an \"optimizer\" to adjust the variables to fit the data.\n", + "\n", + "In this tutorial, we'll walk through a trivial example of a simple linear model: `f(x) = x * W + b`, which has two variables - `W` and `b`. Furthermore, we'll synthesize data such that a well trained model would have `W = 3.0` and `b = 2.0`." + ] + }, + { + "metadata": { + "id": "gFzH64Jn9PIm", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define the model\n", + "\n", + "Let's define a simple class to encapsulate the variables and the computation." + ] + }, + { + "metadata": { + "id": "_WRu7Pze7wk8", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "class Model(object):\n", + " def __init__(self):\n", + " # Initialize variable to (5.0, 0.0)\n", + " # In practice, these should be initialized to random values.\n", + " self.W = tfe.Variable(5.0)\n", + " self.b = tfe.Variable(0.0)\n", + " \n", + " def __call__(self, x):\n", + " return self.W * x + self.b\n", + " \n", + "model = Model()\n", + "\n", + "assert model(3.0).numpy() == 15.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "xa6j_yXa-j79", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define a loss function\n", + "\n", + "A loss function measures how well the output of a model for a given input matches the desired output. Let's use the standard L2 loss." + ] + }, + { + "metadata": { + "id": "Y0ysUFGY924U", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def loss(predicted_y, desired_y):\n", + " return tf.reduce_mean(tf.square(predicted_y - desired_y))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "qutT_fkl_CBc", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Obtain training data\n", + "\n", + "Let's synthesize the training data with some noise." + ] + }, + { + "metadata": { + "id": "gxPTb-kt_N5m", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "TRUE_W = 3.0\n", + "TRUE_b = 2.0\n", + "NUM_EXAMPLES = 1000\n", + "\n", + "inputs = tf.random_normal(shape=[NUM_EXAMPLES])\n", + "noise = tf.random_normal(shape=[NUM_EXAMPLES])\n", + "outputs = inputs * TRUE_W + TRUE_b + noise" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "-50nq-wPBsAW", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Before we train the model let's visualize where the model stands right now. We'll plot the model's predictions in red and the training data in blue." + ] + }, + { + "metadata": { + "id": "_eb83LtrB4nt", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 293 + }, + "outputId": "3873f508-72fb-41e7-a7f5-3f513deefe38", + "executionInfo": { + "elapsed": 1210, + "status": "ok", + "timestamp": 1527005898290, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "import matplotlib.pyplot as plt\n", + "\n", + "plt.scatter(inputs, outputs, c='b')\n", + "plt.scatter(inputs, model(inputs), c='r')\n", + "plt.show()\n", + "\n", + "print('Current loss: '),\n", + "print(loss(model(inputs), outputs).numpy())" + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "display_data", + "data": { + "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEDCAYAAAA2k7/eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\noZkggi8IgtBMEMEXBEFoJvw//5K32R/vBHAAAAAASUVORK5CYII=\n", + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + } + }, + { + "output_type": "stream", + "text": [ + "Current loss: 9.48636\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "sSDP-yeq_4jE", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define a training loop\n", + "\n", + "We now have our network and our training data. Let's train it, i.e., use the training data to update the model's variables (`W` and `b`) so that the loss goes down using [gradient descent](https://en.wikipedia.org/wiki/Gradient_descent). There are many variants of the gradient descent scheme that are captured in `tf.train.Optimizer` implementations. We'd highly recommend using those implementations, but in the spirit of building from first principles, in this particular example we will implement the basic math ourselves." + ] + }, + { + "metadata": { + "id": "MBIACgdnA55X", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def train(model, inputs, outputs, learning_rate):\n", + " with tf.GradientTape() as t:\n", + " current_loss = loss(model(inputs), outputs)\n", + " dW, db = t.gradient(current_loss, [model.W, model.b])\n", + " model.W.assign_sub(learning_rate * dW)\n", + " model.b.assign_sub(learning_rate * db)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "RwWPaJryD2aN", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Finally, let's repeatedly run through the training data and see how `W` and `b` evolve." + ] + }, + { + "metadata": { + "id": "XdfkR223D9dW", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + }, + "height": 446 + }, + "outputId": "c43591ae-d5ac-4f2b-a8e7-bfce607e0919", + "executionInfo": { + "elapsed": 569, + "status": "ok", + "timestamp": 1527005915434, + "user": { + "displayName": "", + "photoUrl": "", + "userId": "" + }, + "user_tz": 420 + } + }, + "cell_type": "code", + "source": [ + "model = Model()\n", + "\n", + "# Collect the history of W-values and b-values to plot later\n", + "Ws, bs = [], []\n", + "epochs = range(10)\n", + "for epoch in epochs:\n", + " Ws.append(model.W.numpy())\n", + " bs.append(model.b.numpy())\n", + " current_loss = loss(model(inputs), outputs)\n", + "\n", + " train(model, inputs, outputs, learning_rate=0.1)\n", + " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", + " (epoch, Ws[-1], bs[-1], current_loss))\n", + "\n", + "# Let's plot it all\n", + "plt.plot(epochs, Ws, 'r',\n", + " epochs, bs, 'b')\n", + "plt.plot([TRUE_W] * len(epochs), 'r--',\n", + " [TRUE_b] * len(epochs), 'b--')\n", + "plt.legend(['W', 'b', 'true W', 'true_b'])\n", + "plt.show()\n", + " " + ], + "execution_count": 0, + "outputs": [ + { + "output_type": "stream", + "text": [ + "Epoch 0: W=5.00 b=0.00, loss=9.48636\n", + "Epoch 1: W=4.58 b=0.42, loss=6.28101\n", + "Epoch 2: W=4.24 b=0.76, loss=4.29357\n", + "Epoch 3: W=3.98 b=1.02, loss=3.06128\n", + "Epoch 4: W=3.78 b=1.23, loss=2.29721\n", + "Epoch 5: W=3.61 b=1.39, loss=1.82345\n", + "Epoch 6: W=3.49 b=1.52, loss=1.52970\n", + "Epoch 7: W=3.38 b=1.62, loss=1.34756\n", + "Epoch 8: W=3.30 b=1.70, loss=1.23463\n", + "Epoch 9: W=3.24 b=1.76, loss=1.16460\n" + ], + "name": "stdout" + }, + { + "output_type": "display_data", + "data": { + "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW0AAAEDCAYAAAD+/1UIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n", + "text/plain": [ + "" + ] + }, + "metadata": { + "tags": [] + } + } + ] + }, + { + "metadata": { + "id": "vPnIVuaSJwWz", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Next Steps\n", + "\n", + "In this tutorial we covered `Variable`s and built and trained a simple linear model using the TensorFlow primitives discussed so far.\n", + "\n", + "In theory, this is pretty much all you need to use TensorFlow for your machine learning research.\n", + "In practice, particularly for neural networks, the higher level APIs like `tf.keras` will be much more convenient since it provides higher level building blocks (called \"layers\"), utilities to save and restore state, a suite of loss functions, a suite of optimization strategies etc. \n", + "\n", + "The [next tutorial](TODO) will cover these higher level APIs." + ] + } + ] +} \ No newline at end of file -- cgit v1.2.3 From 6e5fda7bb9458741843797d9bc98ac3e3295fe30 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 16:46:22 -0700 Subject: Update eager notebook buttons to point to new filename --- .../eager/python/examples/notebooks/automatic_differentiation.ipynb | 4 ++-- .../contrib/eager/python/examples/notebooks/custom_layers.ipynb | 4 ++-- tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb | 4 ++-- tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb | 4 ++-- .../eager/python/examples/notebooks/variables_models_training.ipynb | 4 ++-- 5 files changed, 10 insertions(+), 10 deletions(-) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb index d7bcf749e8..1f9cd7a80b 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb @@ -70,10 +70,10 @@ "cell_type": "markdown", "source": [ "
\n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
" + "View source on GitHub" ] }, { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb index 3378054be4..0c5328f17c 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb @@ -74,10 +74,10 @@ "cell_type": "markdown", "source": [ "
\n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
" + "View source on GitHub" ] }, { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb index 37f2f4f402..88f68dd427 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb @@ -70,10 +70,10 @@ "cell_type": "markdown", "source": [ "
\n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
" + "View source on GitHub" ] }, { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb index 99d141bf2b..b8cfc04ed4 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb @@ -70,10 +70,10 @@ "cell_type": "markdown", "source": [ "
\n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
" + "View source on GitHub" ] }, { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb index ca0262eade..b7e07bbd9f 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb @@ -70,10 +70,10 @@ "cell_type": "markdown", "source": [ "
\n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
" + "View source on GitHub" ] }, { -- cgit v1.2.3 From 8f3dd1da445912e29f0f60b918175c1beda7a798 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 16:56:04 -0700 Subject: Update get_started landing page to point to eager notebook file renames --- tensorflow/docs_src/get_started/_index.yaml | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/tensorflow/docs_src/get_started/_index.yaml b/tensorflow/docs_src/get_started/_index.yaml index 4060804892..07f4ece443 100644 --- a/tensorflow/docs_src/get_started/_index.yaml +++ b/tensorflow/docs_src/get_started/_index.yaml @@ -124,30 +124,30 @@ landing_page:
  1. {% dynamic if request.tld == 'cn' %} - Eager execution basics + Eager execution basics {% dynamic else %} - Eager execution basics + Eager execution basics {% dynamic endif %}
  2. {% dynamic if request.tld == 'cn' %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic else %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic endif %}
  3. {% dynamic if request.tld == 'cn' %} - Variables, models, and training + Variables, models, and training {% dynamic else %} - Variables, models, and training + Variables, models, and training {% dynamic endif %}
  4. {% dynamic if request.tld == 'cn' %} - Custom layers + Custom layers {% dynamic else %} - Custom layers + Custom layers {% dynamic endif %}
  5. Custom training walkthrough
    6. -- cgit v1.2.3 From 6903451efd40a0b8aa5adb324382790f966c273f Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 17:20:37 -0700 Subject: Enable private output on eager notebooks --- .../notebooks/automatic_differentiation.ipynb | 37 +--- .../python/examples/notebooks/custom_layers.ipynb | 242 ++------------------- .../eager/python/examples/notebooks/datasets.ipynb | 40 +--- .../python/examples/notebooks/eager_basics.ipynb | 156 ++----------- .../notebooks/variables_models_training.ipynb | 90 +------- 5 files changed, 58 insertions(+), 507 deletions(-) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb index 1f9cd7a80b..a18882fafa 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/automatic_differentiation.ipynb @@ -3,13 +3,18 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Automatic differentiation and gradient tape", + "name": "automatic_differentiation.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, "provenance": [], + "private_outputs": true, "collapsed_sections": [], "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" } }, "cells": [ @@ -179,20 +184,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 276 - }, - "outputId": "e23f8cc6-6813-4944-f20f-825b8a03c2ff", - "executionInfo": { - "elapsed": 730, - "status": "ok", - "timestamp": 1527005655565, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -215,20 +207,7 @@ "plt.show()" ], "execution_count": 0, - "outputs": [ - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEDCAYAAAAhsS8XAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nF3bGGIswXNgZYyzC/Be68EGj7hfMcwAAAABJRU5ErkJggg==\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - } - ] + "outputs": [] }, { "metadata": { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb index 0c5328f17c..54fbf2a7e1 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/custom_layers.ipynb @@ -3,11 +3,12 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Custom layers", + "name": "custom_layers.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, "provenance": [], + "private_outputs": true, "collapsed_sections": [], "toc_visible": true }, @@ -171,20 +172,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 204 - }, - "outputId": "c5d52434-d980-4488-efa7-5660819d0207", - "executionInfo": { - "elapsed": 244, - "status": "ok", - "timestamp": 1527783641557, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -193,30 +181,7 @@ "layer(tf.zeros([10, 5]))" ], "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 3 - } - ] + "outputs": [] }, { "metadata": { @@ -226,20 +191,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 221 - }, - "outputId": "f0d96dce-0128-4080-bfe2-0ee6fbc0ad90", - "executionInfo": { - "elapsed": 320, - "status": "ok", - "timestamp": 1527783642457, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -250,31 +202,7 @@ "layer.variables" ], "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "[,\n", - " ]" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 4 - } - ] + "outputs": [] }, { "metadata": { @@ -284,20 +212,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 221 - }, - "outputId": "f647fced-c2d7-41a3-c237-242036784665", - "executionInfo": { - "elapsed": 226, - "status": "ok", - "timestamp": 1527783643252, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -306,31 +221,7 @@ "layer.kernel, layer.bias" ], "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "(,\n", - " )" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 5 - } - ] + "outputs": [] }, { "metadata": { @@ -356,20 +247,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 391 - }, - "outputId": "6e7f9285-649a-4132-82ce-73ea92f15862", - "executionInfo": { - "elapsed": 251, - "status": "ok", - "timestamp": 1527783661512, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -392,36 +270,7 @@ "print(layer.variables)" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(\n", - "[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]\n", - " [ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]], shape=(10, 10), dtype=float32)\n", - "[]\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -457,20 +306,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 190 - }, - "outputId": "a8b23a8e-5cf9-4bbf-f93b-6c763d74e2b3", - "executionInfo": { - "elapsed": 420, - "status": "ok", - "timestamp": 1527783698512, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -510,23 +346,7 @@ "print([x.name for x in block.variables])" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(\n", - "[[[[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]\n", - "\n", - " [[ 0. 0. 0.]\n", - " [ 0. 0. 0.]\n", - " [ 0. 0. 0.]]]], shape=(1, 2, 3, 3), dtype=float32)\n", - "['resnet_identity_block_1/conv2d_3/kernel:0', 'resnet_identity_block_1/conv2d_3/bias:0', 'resnet_identity_block_1/batch_normalization_3/gamma:0', 'resnet_identity_block_1/batch_normalization_3/beta:0', 'resnet_identity_block_1/conv2d_4/kernel:0', 'resnet_identity_block_1/conv2d_4/bias:0', 'resnet_identity_block_1/batch_normalization_4/gamma:0', 'resnet_identity_block_1/batch_normalization_4/beta:0', 'resnet_identity_block_1/conv2d_5/kernel:0', 'resnet_identity_block_1/conv2d_5/bias:0', 'resnet_identity_block_1/batch_normalization_5/gamma:0', 'resnet_identity_block_1/batch_normalization_5/beta:0', 'resnet_identity_block_1/batch_normalization_3/moving_mean:0', 'resnet_identity_block_1/batch_normalization_3/moving_variance:0', 'resnet_identity_block_1/batch_normalization_4/moving_mean:0', 'resnet_identity_block_1/batch_normalization_4/moving_variance:0', 'resnet_identity_block_1/batch_normalization_5/moving_mean:0', 'resnet_identity_block_1/batch_normalization_5/moving_variance:0']\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -546,21 +366,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "base_uri": "https://localhost:8080/", - "height": 153 - }, - "outputId": "882e9076-b6d9-4380-bb1e-7c6b57d54c39", - "executionInfo": { - "elapsed": 361, - "status": "ok", - "timestamp": 1526674830777, - "user": { - "displayName": "Alexandre Passos", - "photoUrl": "//lh4.googleusercontent.com/-kmTTWXEgAPw/AAAAAAAAAAI/AAAAAAAAAC0/q_DoOzKGwds/s50-c-k-no/photo.jpg", - "userId": "108023195365833072773" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -575,27 +381,7 @@ "my_seq(tf.zeros([1, 2, 3, 3]))" ], "execution_count": 0, - "outputs": [ - { - "output_type": "execute_result", - "data": { - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - }, - "execution_count": 26 - } - ] + "outputs": [] }, { "metadata": { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb index 88f68dd427..ba6602cb16 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb @@ -3,13 +3,18 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Datasets", + "name": "datasets.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, "provenance": [], + "private_outputs": true, "collapsed_sections": [], "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" } }, "cells": [ @@ -212,21 +217,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "base_uri": "https://localhost:8080/", - "height": 153 - }, - "outputId": "8e4b0298-d27d-4ac7-e26a-ef94af0594ec", - "executionInfo": { - "elapsed": 388, - "status": "ok", - "timestamp": 1525154629129, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -240,22 +231,7 @@ " print(x)" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Elements of ds_tensors:\n", - "tf.Tensor([1 9], shape=(2,), dtype=int32)\n", - "tf.Tensor([16 25], shape=(2,), dtype=int32)\n", - "tf.Tensor([ 4 36], shape=(2,), dtype=int32)\n", - "\n", - "Elements in ds_file:\n", - "tf.Tensor(['Line 1' 'Line 2'], shape=(2,), dtype=string)\n", - "tf.Tensor(['Line 3' ' '], shape=(2,), dtype=string)\n" - ], - "name": "stdout" - } - ] + "outputs": [] } ] } \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb index b8cfc04ed4..75079b8b7e 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb @@ -3,13 +3,18 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Eager execution basics", + "name": "eager_basics.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, "provenance": [], + "private_outputs": true, "collapsed_sections": [], "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" } }, "cells": [ @@ -144,22 +149,9 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 125 + } }, - "cellView": "code", - "outputId": "b1a1cd60-4eb3-443d-cd6b-68406390784e", - "executionInfo": { - "elapsed": 320, - "status": "ok", - "timestamp": 1526420535530, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } + "cellView": "code" }, "cell_type": "code", "source": [ @@ -173,20 +165,7 @@ "print(tf.square(2) + tf.square(3))" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "tf.Tensor(3, shape=(), dtype=int32)\n", - "tf.Tensor([4 6], shape=(2,), dtype=int32)\n", - "tf.Tensor(25, shape=(), dtype=int32)\n", - "tf.Tensor(6, shape=(), dtype=int32)\n", - "tf.Tensor(aGVsbG8gd29ybGQ, shape=(), dtype=string)\n", - "tf.Tensor(13, shape=(), dtype=int32)\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -206,20 +185,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 53 - }, - "outputId": "5e4ac41c-5115-4e50-eba0-42e249c16561", - "executionInfo": { - "elapsed": 215, - "status": "ok", - "timestamp": 1526420538162, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -229,16 +195,7 @@ "print(x.dtype)" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "(1, 2)\n", - "\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -278,20 +235,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 251 - }, - "outputId": "fd0a22bc-8249-49dd-fcbd-63161cc47e46", - "executionInfo": { - "elapsed": 238, - "status": "ok", - "timestamp": 1526420540562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -312,27 +256,7 @@ "print(tensor.numpy())" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "TensorFlow operations convert numpy arrays to Tensors automatically\n", - "tf.Tensor(\n", - "[[ 42. 42. 42.]\n", - " [ 42. 42. 42.]\n", - " [ 42. 42. 42.]], shape=(3, 3), dtype=float64)\n", - "And NumPy operations convert Tensors to numpy arrays automatically\n", - "[[ 43. 43. 43.]\n", - " [ 43. 43. 43.]\n", - " [ 43. 43. 43.]]\n", - "The .numpy() method explicitly converts a Tensor to a numpy array\n", - "[[ 42. 42. 42.]\n", - " [ 42. 42. 42.]\n", - " [ 42. 42. 42.]]\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -354,22 +278,9 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 53 + } }, - "cellView": "code", - "outputId": "2239ae2b-adf3-4895-b1f3-464cf5361d1b", - "executionInfo": { - "elapsed": 340, - "status": "ok", - "timestamp": 1526420543562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 - } + "cellView": "code" }, "cell_type": "code", "source": [ @@ -382,16 +293,7 @@ "print(x.device.endswith('GPU:0'))" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Is there a GPU available: False\n", - "Is the Tensor on GPU #0: False\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -427,20 +329,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 53 - }, - "outputId": "2e613293-ccac-4db2-b793-8ceb5b5adcfd", - "executionInfo": { - "elapsed": 1762, - "status": "ok", - "timestamp": 1526420547562, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -463,16 +352,7 @@ " time_matmul(x)" ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "On CPU:\n", - "10 loops, best of 3: 35.8 ms per loop\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb index b7e07bbd9f..28f91dc11a 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb @@ -3,13 +3,18 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Variables, models, and training", + "name": "variables_models_training.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, "provenance": [], + "private_outputs": true, "collapsed_sections": [], "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" } }, "cells": [ @@ -350,20 +355,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 293 - }, - "outputId": "3873f508-72fb-41e7-a7f5-3f513deefe38", - "executionInfo": { - "elapsed": 1210, - "status": "ok", - "timestamp": 1527005898290, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -378,27 +370,7 @@ "print(loss(model(inputs), outputs).numpy())" ], "execution_count": 0, - "outputs": [ - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEDCAYAAAA2k7/eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\noZkggi8IgtBMEMEXBEFoJvw//5K32R/vBHAAAAAASUVORK5CYII=\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - }, - { - "output_type": "stream", - "text": [ - "Current loss: 9.48636\n" - ], - "name": "stdout" - } - ] + "outputs": [] }, { "metadata": { @@ -453,20 +425,7 @@ "autoexec": { "startup": false, "wait_interval": 0 - }, - "height": 446 - }, - "outputId": "c43591ae-d5ac-4f2b-a8e7-bfce607e0919", - "executionInfo": { - "elapsed": 569, - "status": "ok", - "timestamp": 1527005915434, - "user": { - "displayName": "", - "photoUrl": "", - "userId": "" - }, - "user_tz": 420 + } } }, "cell_type": "code", @@ -495,36 +454,7 @@ " " ], "execution_count": 0, - "outputs": [ - { - "output_type": "stream", - "text": [ - "Epoch 0: W=5.00 b=0.00, loss=9.48636\n", - "Epoch 1: W=4.58 b=0.42, loss=6.28101\n", - "Epoch 2: W=4.24 b=0.76, loss=4.29357\n", - "Epoch 3: W=3.98 b=1.02, loss=3.06128\n", - "Epoch 4: W=3.78 b=1.23, loss=2.29721\n", - "Epoch 5: W=3.61 b=1.39, loss=1.82345\n", - "Epoch 6: W=3.49 b=1.52, loss=1.52970\n", - "Epoch 7: W=3.38 b=1.62, loss=1.34756\n", - "Epoch 8: W=3.30 b=1.70, loss=1.23463\n", - "Epoch 9: W=3.24 b=1.76, loss=1.16460\n" - ], - "name": "stdout" - }, - { - "output_type": "display_data", - "data": { - "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW0AAAEDCAYAAAD+/1UIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n", - "text/plain": [ - "" - ] - }, - "metadata": { - "tags": [] - } - } - ] + "outputs": [] }, { "metadata": { -- cgit v1.2.3 From d006cea6971c03e8c392b498440c937b6f3863e1 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 17:25:12 -0700 Subject: Rename eager notebook to Custom Training --- .../examples/notebooks/variables_models_training.ipynb | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb index 28f91dc11a..d5b9e6ee48 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb @@ -3,7 +3,7 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "variables_models_training.ipynb", + "name": "Custom training: variables and models", "version": "0.3.2", "views": {}, "default_view": {}, @@ -64,7 +64,7 @@ }, "cell_type": "markdown", "source": [ - "# Variables, models, and training" + "# Custom training: variables and models" ] }, { @@ -75,10 +75,10 @@ "cell_type": "markdown", "source": [ "
    \n", - "\n", + "\n", " Run in Google Colab\n", "\n", - "View source on GitHub
    " + "View source on GitHub" ] }, { @@ -118,8 +118,9 @@ "cell_type": "code", "source": [ "import tensorflow as tf\n", - "tf.enable_eager_execution()\n", - "tfe = tf.contrib.eager # Shorthand for some symbols" + "tfe = tf.contrib.eager # Shorthand for some symbols\n", + "\n", + "tf.enable_eager_execution()" ], "execution_count": 0, "outputs": [] -- cgit v1.2.3 From 7e6d0cb59f3823f50c6003a10eb507ca0579a095 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 17:28:05 -0700 Subject: Filename change for eager custom training notebook --- .../eager/python/examples/notebooks/README.md | 2 +- .../examples/notebooks/custom_training.ipynb | 478 +++++++++++++++++++++ .../notebooks/variables_models_training.ipynb | 478 --------------------- tensorflow/docs_src/get_started/_index.yaml | 4 +- 4 files changed, 481 insertions(+), 481 deletions(-) create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb diff --git a/tensorflow/contrib/eager/python/examples/notebooks/README.md b/tensorflow/contrib/eager/python/examples/notebooks/README.md index 2a833bc683..b7a2133d9c 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/README.md +++ b/tensorflow/contrib/eager/python/examples/notebooks/README.md @@ -7,5 +7,5 @@ differentiation. Start with these notebooks, then read the 1. [Eager execution basics](./eager_basics.ipynb) 2. [Automatic differentiation and gradient tapes](./automatic_differentiation.ipynb) -3. [Variables, models, and training](./variables_models_training.ipynb) +3. [Custom training: variables and models](./custom_training.ipynb) 4. [Custom layers](./custom_layers.ipynb) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb new file mode 100644 index 0000000000..d5b9e6ee48 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb @@ -0,0 +1,478 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Custom training: variables and models", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "private_outputs": true, + "collapsed_sections": [], + "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" + } + }, + "cells": [ + { + "metadata": { + "id": "5rmpybwysXGV", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "m8y3rGtQsYP2", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "hrXv0rU9sIma", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Custom training: variables and models" + ] + }, + { + "metadata": { + "id": "7S0BwJ_8sLu7", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
    \n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
    " + ] + }, + { + "metadata": { + "id": "k2o3TTG4TFpt", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "In the previous tutorial we covered the TensorFlow APIs for automatic differentiation, a basic building block for machine learning.\n", + "In this tutorial we will use the TensorFlow primitives introduced in the prior tutorials to do some simple machine learning.\n", + "\n", + "TensorFlow also includes a higher-level neural networks API (`tf.keras`) which provides useful abstractions to reduce boilerplate. We strongly recommend those higher level APIs for people working with neural networks. However, in this short tutorial we cover neural network training from first principles to establish a strong foundation." + ] + }, + { + "metadata": { + "id": "3LXMVuV0VhDr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Setup" + ] + }, + { + "metadata": { + "id": "PJ64L90aVir3", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "tfe = tf.contrib.eager # Shorthand for some symbols\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "eMAWbDJFVmMk", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Variables\n", + "\n", + "Tensors in TensorFlow are immutable stateless objects. Machine learning models, however, need to have changing state: as your model trains, the same code to compute predictions should behave differently over time (hopefully with a lower loss!). To represent this state which needs to change over the course of your computation, you can choose to rely on the fact that Python is a stateful programming language:\n" + ] + }, + { + "metadata": { + "id": "VkJwtLS_Jbn8", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "# Using python state\n", + "x = tf.zeros([10, 10])\n", + "x += 2 # This is equivalent to x = x + 2, which does not mutate the original\n", + " # value of x\n", + "print(x)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "wfneTXy7JcUz", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "TensorFlow, however, has stateful operations built in, and these are often more pleasant to use than low-level Python representations of your state. To represent weights in a model, for example, it's often convenient and efficient to use TensorFlow variables.\n", + "\n", + "A Variable is an object which stores a value and, when used in a TensorFlow computation, will implicitly read from this stored value. There are operations (`tf.assign_sub`, `tf.scatter_update`, etc) which manipulate the value stored in a TensorFlow variable." + ] + }, + { + "metadata": { + "id": "itxmrMil6DQi", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "v = tfe.Variable(1.0)\n", + "assert v.numpy() == 1.0\n", + "\n", + "# Re-assign the value\n", + "v.assign(3.0)\n", + "assert v.numpy() == 3.0\n", + "\n", + "# Use `v` in a TensorFlow operation like tf.square() and reassign\n", + "v.assign(tf.square(v))\n", + "assert v.numpy() == 9.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "-paSaeq1JzwC", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Computations using Variables are automatically traced when computing gradients. For Variables representing embeddings TensorFlow will do sparse updates by default, which are more computation and memory efficient.\n", + "\n", + "Using Variables is also a way to quickly let a reader of your code know that this piece of state is mutable." + ] + }, + { + "metadata": { + "id": "BMiFcDzE7Qu3", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Example: Fitting a linear model\n", + "\n", + "Let's now put the few concepts we have so far ---`Tensor`, `GradientTape`, `Variable` --- to build and train a simple model. This typically involves a few steps:\n", + "\n", + "1. Define the model.\n", + "2. Define a loss function.\n", + "3. Obtain training data.\n", + "4. Run through the training data and use an \"optimizer\" to adjust the variables to fit the data.\n", + "\n", + "In this tutorial, we'll walk through a trivial example of a simple linear model: `f(x) = x * W + b`, which has two variables - `W` and `b`. Furthermore, we'll synthesize data such that a well trained model would have `W = 3.0` and `b = 2.0`." + ] + }, + { + "metadata": { + "id": "gFzH64Jn9PIm", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define the model\n", + "\n", + "Let's define a simple class to encapsulate the variables and the computation." + ] + }, + { + "metadata": { + "id": "_WRu7Pze7wk8", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "class Model(object):\n", + " def __init__(self):\n", + " # Initialize variable to (5.0, 0.0)\n", + " # In practice, these should be initialized to random values.\n", + " self.W = tfe.Variable(5.0)\n", + " self.b = tfe.Variable(0.0)\n", + " \n", + " def __call__(self, x):\n", + " return self.W * x + self.b\n", + " \n", + "model = Model()\n", + "\n", + "assert model(3.0).numpy() == 15.0" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "xa6j_yXa-j79", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define a loss function\n", + "\n", + "A loss function measures how well the output of a model for a given input matches the desired output. Let's use the standard L2 loss." + ] + }, + { + "metadata": { + "id": "Y0ysUFGY924U", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def loss(predicted_y, desired_y):\n", + " return tf.reduce_mean(tf.square(predicted_y - desired_y))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "qutT_fkl_CBc", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Obtain training data\n", + "\n", + "Let's synthesize the training data with some noise." + ] + }, + { + "metadata": { + "id": "gxPTb-kt_N5m", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "TRUE_W = 3.0\n", + "TRUE_b = 2.0\n", + "NUM_EXAMPLES = 1000\n", + "\n", + "inputs = tf.random_normal(shape=[NUM_EXAMPLES])\n", + "noise = tf.random_normal(shape=[NUM_EXAMPLES])\n", + "outputs = inputs * TRUE_W + TRUE_b + noise" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "-50nq-wPBsAW", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Before we train the model let's visualize where the model stands right now. We'll plot the model's predictions in red and the training data in blue." + ] + }, + { + "metadata": { + "id": "_eb83LtrB4nt", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import matplotlib.pyplot as plt\n", + "\n", + "plt.scatter(inputs, outputs, c='b')\n", + "plt.scatter(inputs, model(inputs), c='r')\n", + "plt.show()\n", + "\n", + "print('Current loss: '),\n", + "print(loss(model(inputs), outputs).numpy())" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "sSDP-yeq_4jE", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Define a training loop\n", + "\n", + "We now have our network and our training data. Let's train it, i.e., use the training data to update the model's variables (`W` and `b`) so that the loss goes down using [gradient descent](https://en.wikipedia.org/wiki/Gradient_descent). There are many variants of the gradient descent scheme that are captured in `tf.train.Optimizer` implementations. We'd highly recommend using those implementations, but in the spirit of building from first principles, in this particular example we will implement the basic math ourselves." + ] + }, + { + "metadata": { + "id": "MBIACgdnA55X", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def train(model, inputs, outputs, learning_rate):\n", + " with tf.GradientTape() as t:\n", + " current_loss = loss(model(inputs), outputs)\n", + " dW, db = t.gradient(current_loss, [model.W, model.b])\n", + " model.W.assign_sub(learning_rate * dW)\n", + " model.b.assign_sub(learning_rate * db)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "RwWPaJryD2aN", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Finally, let's repeatedly run through the training data and see how `W` and `b` evolve." + ] + }, + { + "metadata": { + "id": "XdfkR223D9dW", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "model = Model()\n", + "\n", + "# Collect the history of W-values and b-values to plot later\n", + "Ws, bs = [], []\n", + "epochs = range(10)\n", + "for epoch in epochs:\n", + " Ws.append(model.W.numpy())\n", + " bs.append(model.b.numpy())\n", + " current_loss = loss(model(inputs), outputs)\n", + "\n", + " train(model, inputs, outputs, learning_rate=0.1)\n", + " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", + " (epoch, Ws[-1], bs[-1], current_loss))\n", + "\n", + "# Let's plot it all\n", + "plt.plot(epochs, Ws, 'r',\n", + " epochs, bs, 'b')\n", + "plt.plot([TRUE_W] * len(epochs), 'r--',\n", + " [TRUE_b] * len(epochs), 'b--')\n", + "plt.legend(['W', 'b', 'true W', 'true_b'])\n", + "plt.show()\n", + " " + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "vPnIVuaSJwWz", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Next Steps\n", + "\n", + "In this tutorial we covered `Variable`s and built and trained a simple linear model using the TensorFlow primitives discussed so far.\n", + "\n", + "In theory, this is pretty much all you need to use TensorFlow for your machine learning research.\n", + "In practice, particularly for neural networks, the higher level APIs like `tf.keras` will be much more convenient since it provides higher level building blocks (called \"layers\"), utilities to save and restore state, a suite of loss functions, a suite of optimization strategies etc. \n", + "\n", + "The [next tutorial](TODO) will cover these higher level APIs." + ] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb deleted file mode 100644 index d5b9e6ee48..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/variables_models_training.ipynb +++ /dev/null @@ -1,478 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "Custom training: variables and models", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "private_outputs": true, - "collapsed_sections": [], - "toc_visible": true - }, - "kernelspec": { - "name": "python3", - "display_name": "Python 3" - } - }, - "cells": [ - { - "metadata": { - "id": "5rmpybwysXGV", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "m8y3rGtQsYP2", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "hrXv0rU9sIma", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Custom training: variables and models" - ] - }, - { - "metadata": { - "id": "7S0BwJ_8sLu7", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
    \n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
    " - ] - }, - { - "metadata": { - "id": "k2o3TTG4TFpt", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "In the previous tutorial we covered the TensorFlow APIs for automatic differentiation, a basic building block for machine learning.\n", - "In this tutorial we will use the TensorFlow primitives introduced in the prior tutorials to do some simple machine learning.\n", - "\n", - "TensorFlow also includes a higher-level neural networks API (`tf.keras`) which provides useful abstractions to reduce boilerplate. We strongly recommend those higher level APIs for people working with neural networks. However, in this short tutorial we cover neural network training from first principles to establish a strong foundation." - ] - }, - { - "metadata": { - "id": "3LXMVuV0VhDr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Setup" - ] - }, - { - "metadata": { - "id": "PJ64L90aVir3", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "tfe = tf.contrib.eager # Shorthand for some symbols\n", - "\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "eMAWbDJFVmMk", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Variables\n", - "\n", - "Tensors in TensorFlow are immutable stateless objects. Machine learning models, however, need to have changing state: as your model trains, the same code to compute predictions should behave differently over time (hopefully with a lower loss!). To represent this state which needs to change over the course of your computation, you can choose to rely on the fact that Python is a stateful programming language:\n" - ] - }, - { - "metadata": { - "id": "VkJwtLS_Jbn8", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "# Using python state\n", - "x = tf.zeros([10, 10])\n", - "x += 2 # This is equivalent to x = x + 2, which does not mutate the original\n", - " # value of x\n", - "print(x)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "wfneTXy7JcUz", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "TensorFlow, however, has stateful operations built in, and these are often more pleasant to use than low-level Python representations of your state. To represent weights in a model, for example, it's often convenient and efficient to use TensorFlow variables.\n", - "\n", - "A Variable is an object which stores a value and, when used in a TensorFlow computation, will implicitly read from this stored value. There are operations (`tf.assign_sub`, `tf.scatter_update`, etc) which manipulate the value stored in a TensorFlow variable." - ] - }, - { - "metadata": { - "id": "itxmrMil6DQi", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "v = tfe.Variable(1.0)\n", - "assert v.numpy() == 1.0\n", - "\n", - "# Re-assign the value\n", - "v.assign(3.0)\n", - "assert v.numpy() == 3.0\n", - "\n", - "# Use `v` in a TensorFlow operation like tf.square() and reassign\n", - "v.assign(tf.square(v))\n", - "assert v.numpy() == 9.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "-paSaeq1JzwC", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Computations using Variables are automatically traced when computing gradients. For Variables representing embeddings TensorFlow will do sparse updates by default, which are more computation and memory efficient.\n", - "\n", - "Using Variables is also a way to quickly let a reader of your code know that this piece of state is mutable." - ] - }, - { - "metadata": { - "id": "BMiFcDzE7Qu3", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Example: Fitting a linear model\n", - "\n", - "Let's now put the few concepts we have so far ---`Tensor`, `GradientTape`, `Variable` --- to build and train a simple model. This typically involves a few steps:\n", - "\n", - "1. Define the model.\n", - "2. Define a loss function.\n", - "3. Obtain training data.\n", - "4. Run through the training data and use an \"optimizer\" to adjust the variables to fit the data.\n", - "\n", - "In this tutorial, we'll walk through a trivial example of a simple linear model: `f(x) = x * W + b`, which has two variables - `W` and `b`. Furthermore, we'll synthesize data such that a well trained model would have `W = 3.0` and `b = 2.0`." - ] - }, - { - "metadata": { - "id": "gFzH64Jn9PIm", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define the model\n", - "\n", - "Let's define a simple class to encapsulate the variables and the computation." - ] - }, - { - "metadata": { - "id": "_WRu7Pze7wk8", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "class Model(object):\n", - " def __init__(self):\n", - " # Initialize variable to (5.0, 0.0)\n", - " # In practice, these should be initialized to random values.\n", - " self.W = tfe.Variable(5.0)\n", - " self.b = tfe.Variable(0.0)\n", - " \n", - " def __call__(self, x):\n", - " return self.W * x + self.b\n", - " \n", - "model = Model()\n", - "\n", - "assert model(3.0).numpy() == 15.0" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "xa6j_yXa-j79", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define a loss function\n", - "\n", - "A loss function measures how well the output of a model for a given input matches the desired output. Let's use the standard L2 loss." - ] - }, - { - "metadata": { - "id": "Y0ysUFGY924U", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def loss(predicted_y, desired_y):\n", - " return tf.reduce_mean(tf.square(predicted_y - desired_y))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "qutT_fkl_CBc", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Obtain training data\n", - "\n", - "Let's synthesize the training data with some noise." - ] - }, - { - "metadata": { - "id": "gxPTb-kt_N5m", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "TRUE_W = 3.0\n", - "TRUE_b = 2.0\n", - "NUM_EXAMPLES = 1000\n", - "\n", - "inputs = tf.random_normal(shape=[NUM_EXAMPLES])\n", - "noise = tf.random_normal(shape=[NUM_EXAMPLES])\n", - "outputs = inputs * TRUE_W + TRUE_b + noise" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "-50nq-wPBsAW", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Before we train the model let's visualize where the model stands right now. We'll plot the model's predictions in red and the training data in blue." - ] - }, - { - "metadata": { - "id": "_eb83LtrB4nt", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import matplotlib.pyplot as plt\n", - "\n", - "plt.scatter(inputs, outputs, c='b')\n", - "plt.scatter(inputs, model(inputs), c='r')\n", - "plt.show()\n", - "\n", - "print('Current loss: '),\n", - "print(loss(model(inputs), outputs).numpy())" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "sSDP-yeq_4jE", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Define a training loop\n", - "\n", - "We now have our network and our training data. Let's train it, i.e., use the training data to update the model's variables (`W` and `b`) so that the loss goes down using [gradient descent](https://en.wikipedia.org/wiki/Gradient_descent). There are many variants of the gradient descent scheme that are captured in `tf.train.Optimizer` implementations. We'd highly recommend using those implementations, but in the spirit of building from first principles, in this particular example we will implement the basic math ourselves." - ] - }, - { - "metadata": { - "id": "MBIACgdnA55X", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def train(model, inputs, outputs, learning_rate):\n", - " with tf.GradientTape() as t:\n", - " current_loss = loss(model(inputs), outputs)\n", - " dW, db = t.gradient(current_loss, [model.W, model.b])\n", - " model.W.assign_sub(learning_rate * dW)\n", - " model.b.assign_sub(learning_rate * db)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "RwWPaJryD2aN", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Finally, let's repeatedly run through the training data and see how `W` and `b` evolve." - ] - }, - { - "metadata": { - "id": "XdfkR223D9dW", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "model = Model()\n", - "\n", - "# Collect the history of W-values and b-values to plot later\n", - "Ws, bs = [], []\n", - "epochs = range(10)\n", - "for epoch in epochs:\n", - " Ws.append(model.W.numpy())\n", - " bs.append(model.b.numpy())\n", - " current_loss = loss(model(inputs), outputs)\n", - "\n", - " train(model, inputs, outputs, learning_rate=0.1)\n", - " print('Epoch %2d: W=%1.2f b=%1.2f, loss=%2.5f' %\n", - " (epoch, Ws[-1], bs[-1], current_loss))\n", - "\n", - "# Let's plot it all\n", - "plt.plot(epochs, Ws, 'r',\n", - " epochs, bs, 'b')\n", - "plt.plot([TRUE_W] * len(epochs), 'r--',\n", - " [TRUE_b] * len(epochs), 'b--')\n", - "plt.legend(['W', 'b', 'true W', 'true_b'])\n", - "plt.show()\n", - " " - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "vPnIVuaSJwWz", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Next Steps\n", - "\n", - "In this tutorial we covered `Variable`s and built and trained a simple linear model using the TensorFlow primitives discussed so far.\n", - "\n", - "In theory, this is pretty much all you need to use TensorFlow for your machine learning research.\n", - "In practice, particularly for neural networks, the higher level APIs like `tf.keras` will be much more convenient since it provides higher level building blocks (called \"layers\"), utilities to save and restore state, a suite of loss functions, a suite of optimization strategies etc. \n", - "\n", - "The [next tutorial](TODO) will cover these higher level APIs." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/docs_src/get_started/_index.yaml b/tensorflow/docs_src/get_started/_index.yaml index 07f4ece443..c3e216b2a6 100644 --- a/tensorflow/docs_src/get_started/_index.yaml +++ b/tensorflow/docs_src/get_started/_index.yaml @@ -138,9 +138,9 @@ landing_page:
  6. {% dynamic if request.tld == 'cn' %} - Variables, models, and training + Custom training: variables and models {% dynamic else %} - Variables, models, and training + Custom training: variables and models {% dynamic endif %}
  7. -- cgit v1.2.3 From 47d4d0ab70c98d8935d9abc4f4829ded2cd3d8c0 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 22:55:31 -0700 Subject: Rename eager_basics to eager_intro --- .../eager/python/examples/notebooks/README.md | 4 +- .../examples/notebooks/custom_training.ipynb | 4 +- .../python/examples/notebooks/eager_basics.ipynb | 371 --------------------- .../python/examples/notebooks/eager_intro.ipynb | 371 +++++++++++++++++++++ 4 files changed, 375 insertions(+), 375 deletions(-) delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb diff --git a/tensorflow/contrib/eager/python/examples/notebooks/README.md b/tensorflow/contrib/eager/python/examples/notebooks/README.md index b7a2133d9c..c17048d6ea 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/README.md +++ b/tensorflow/contrib/eager/python/examples/notebooks/README.md @@ -5,7 +5,7 @@ operations. Write custom layers, forward passes, and training loops with auto differentiation. Start with these notebooks, then read the [eager execution guide](https://www.tensorflow.org/guide/eager). -1. [Eager execution basics](./eager_basics.ipynb) +1. [Eager execution introduction](./eager_intro.ipynb) 2. [Automatic differentiation and gradient tapes](./automatic_differentiation.ipynb) -3. [Custom training: variables and models](./custom_training.ipynb) +3. [Custom training: basics](./custom_training.ipynb) 4. [Custom layers](./custom_layers.ipynb) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb index d5b9e6ee48..0a781d2153 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/custom_training.ipynb @@ -3,7 +3,7 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Custom training: variables and models", + "name": "Custom training: basics", "version": "0.3.2", "views": {}, "default_view": {}, @@ -64,7 +64,7 @@ }, "cell_type": "markdown", "source": [ - "# Custom training: variables and models" + "# Custom training: basics" ] }, { diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb deleted file mode 100644 index 75079b8b7e..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb +++ /dev/null @@ -1,371 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "eager_basics.ipynb", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "private_outputs": true, - "collapsed_sections": [], - "toc_visible": true - }, - "kernelspec": { - "name": "python3", - "display_name": "Python 3" - } - }, - "cells": [ - { - "metadata": { - "id": "iPpI7RaYoZuE", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "hro2InpHobKk", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "U9i2Dsh-ziXr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Eager execution basics" - ] - }, - { - "metadata": { - "id": "Hndw-YcxoOJK", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
    \n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
    " - ] - }, - { - "metadata": { - "id": "6sILUVbHoSgH", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "This is an introductory tutorial for using TensorFlow. It will cover:\n", - "\n", - "* Importing required packages\n", - "* Creating and using Tensors\n", - "* Using GPU acceleration" - ] - }, - { - "metadata": { - "id": "z1JcS5iBXMRO", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Import TensorFlow\n", - "\n", - "To get started, import the `tensorflow` module and enable eager execution.\n", - "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." - ] - }, - { - "metadata": { - "id": "RlIWhyeLoYnG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "H9UySOPLXdaw", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Tensors\n", - "\n", - "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" - ] - }, - { - "metadata": { - "id": "ngUe237Wt48W", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "print(tf.add(1, 2))\n", - "print(tf.add([1, 2], [3, 4]))\n", - "print(tf.square(5))\n", - "print(tf.reduce_sum([1, 2, 3]))\n", - "print(tf.encode_base64(\"hello world\"))\n", - "\n", - "# Operator overloading is also supported\n", - "print(tf.square(2) + tf.square(3))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "IDY4WsYRhP81", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Each Tensor has a shape and a datatype" - ] - }, - { - "metadata": { - "id": "srYWH1MdJNG7", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "x = tf.matmul([[1]], [[2, 3]])\n", - "print(x.shape)\n", - "print(x.dtype)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "eBPw8e8vrsom", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", - "\n", - "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", - "2. Tensors are immutable." - ] - }, - { - "metadata": { - "id": "Dwi1tdW3JBw6", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### NumPy Compatibility\n", - "\n", - "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", - "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", - "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", - "\n", - "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", - "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." - ] - }, - { - "metadata": { - "id": "lCUWzso6mbqR", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import numpy as np\n", - "\n", - "ndarray = np.ones([3, 3])\n", - "\n", - "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", - "tensor = tf.multiply(ndarray, 42)\n", - "print(tensor)\n", - "\n", - "\n", - "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", - "print(np.add(tensor, 1))\n", - "\n", - "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", - "print(tensor.numpy())" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "PBNP8yTRfu_X", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## GPU acceleration\n", - "\n", - "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" - ] - }, - { - "metadata": { - "id": "3Twf_Rw-gQFM", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "x = tf.random_uniform([3, 3])\n", - "\n", - "print(\"Is there a GPU available: \"),\n", - "print(tf.test.is_gpu_available())\n", - "\n", - "print(\"Is the Tensor on GPU #0: \"),\n", - "print(x.device.endswith('GPU:0'))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "vpgYzgVXW2Ud", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Device Names\n", - "\n", - "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." - ] - }, - { - "metadata": { - "id": "ZWZQCimzuqyP", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "\n", - "\n", - "### Explicit Device Placement\n", - "\n", - "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" - ] - }, - { - "metadata": { - "id": "RjkNZTuauy-Q", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def time_matmul(x):\n", - " %timeit tf.matmul(x, x)\n", - "\n", - "# Force execution on CPU\n", - "print(\"On CPU:\")\n", - "with tf.device(\"CPU:0\"):\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"CPU:0\")\n", - " time_matmul(x)\n", - "\n", - "# Force execution on GPU #0 if available\n", - "if tf.test.is_gpu_available():\n", - " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"GPU:0\")\n", - " time_matmul(x)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "YEOJTNiOvnpQ", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Next Steps\n", - "\n", - "In this tutorial we covered the most fundamental concepts in TensorFlow - `Tensor`s, operations, and devices.\n", - "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." - ] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb new file mode 100644 index 0000000000..a00893bc77 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb @@ -0,0 +1,371 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "Eager execution introduction", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "private_outputs": true, + "collapsed_sections": [], + "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" + } + }, + "cells": [ + { + "metadata": { + "id": "iPpI7RaYoZuE", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "hro2InpHobKk", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Eager execution introduction" + ] + }, + { + "metadata": { + "id": "Hndw-YcxoOJK", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
    \n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
    " + ] + }, + { + "metadata": { + "id": "6sILUVbHoSgH", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "This is an introductory tutorial for using TensorFlow. It will cover:\n", + "\n", + "* Importing required packages\n", + "* Creating and using Tensors\n", + "* Using GPU acceleration" + ] + }, + { + "metadata": { + "id": "z1JcS5iBXMRO", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Import TensorFlow\n", + "\n", + "To get started, import the `tensorflow` module and enable eager execution.\n", + "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." + ] + }, + { + "metadata": { + "id": "RlIWhyeLoYnG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "H9UySOPLXdaw", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Tensors\n", + "\n", + "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" + ] + }, + { + "metadata": { + "id": "ngUe237Wt48W", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "print(tf.add(1, 2))\n", + "print(tf.add([1, 2], [3, 4]))\n", + "print(tf.square(5))\n", + "print(tf.reduce_sum([1, 2, 3]))\n", + "print(tf.encode_base64(\"hello world\"))\n", + "\n", + "# Operator overloading is also supported\n", + "print(tf.square(2) + tf.square(3))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "IDY4WsYRhP81", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Each Tensor has a shape and a datatype" + ] + }, + { + "metadata": { + "id": "srYWH1MdJNG7", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "x = tf.matmul([[1]], [[2, 3]])\n", + "print(x.shape)\n", + "print(x.dtype)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "eBPw8e8vrsom", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", + "\n", + "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", + "2. Tensors are immutable." + ] + }, + { + "metadata": { + "id": "Dwi1tdW3JBw6", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### NumPy Compatibility\n", + "\n", + "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", + "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", + "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", + "\n", + "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", + "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." + ] + }, + { + "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import numpy as np\n", + "\n", + "ndarray = np.ones([3, 3])\n", + "\n", + "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", + "tensor = tf.multiply(ndarray, 42)\n", + "print(tensor)\n", + "\n", + "\n", + "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", + "print(np.add(tensor, 1))\n", + "\n", + "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", + "print(tensor.numpy())" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "PBNP8yTRfu_X", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## GPU acceleration\n", + "\n", + "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" + ] + }, + { + "metadata": { + "id": "3Twf_Rw-gQFM", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "x = tf.random_uniform([3, 3])\n", + "\n", + "print(\"Is there a GPU available: \"),\n", + "print(tf.test.is_gpu_available())\n", + "\n", + "print(\"Is the Tensor on GPU #0: \"),\n", + "print(x.device.endswith('GPU:0'))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "vpgYzgVXW2Ud", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Device Names\n", + "\n", + "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." + ] + }, + { + "metadata": { + "id": "ZWZQCimzuqyP", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "\n", + "\n", + "### Explicit Device Placement\n", + "\n", + "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" + ] + }, + { + "metadata": { + "id": "RjkNZTuauy-Q", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def time_matmul(x):\n", + " %timeit tf.matmul(x, x)\n", + "\n", + "# Force execution on CPU\n", + "print(\"On CPU:\")\n", + "with tf.device(\"CPU:0\"):\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"CPU:0\")\n", + " time_matmul(x)\n", + "\n", + "# Force execution on GPU #0 if available\n", + "if tf.test.is_gpu_available():\n", + " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"GPU:0\")\n", + " time_matmul(x)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "YEOJTNiOvnpQ", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Next Steps\n", + "\n", + "In this tutorial we covered the most fundamental concepts in TensorFlow - `Tensor`s, operations, and devices.\n", + "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." + ] + } + ] +} \ No newline at end of file -- cgit v1.2.3 From 152b830b379e7d4ec0db9aac111a732db37e19c3 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 23:11:45 -0700 Subject: Move datasets notebook as section in eager_intro --- .../eager/python/examples/notebooks/datasets.ipynb | 237 --------------------- .../python/examples/notebooks/eager_intro.ipynb | 132 +++++++++++- 2 files changed, 126 insertions(+), 243 deletions(-) delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb diff --git a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb deleted file mode 100644 index ba6602cb16..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/datasets.ipynb +++ /dev/null @@ -1,237 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "datasets.ipynb", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "private_outputs": true, - "collapsed_sections": [], - "toc_visible": true - }, - "kernelspec": { - "name": "python3", - "display_name": "Python 3" - } - }, - "cells": [ - { - "metadata": { - "id": "hvpDiXDuuNEu", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "4FWzUGc7uQKr", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "RRbeuncot80C", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Datasets" - ] - }, - { - "metadata": { - "id": "i9rtB6DnuFw5", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
    \n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
    " - ] - }, - { - "metadata": { - "id": "U9i2Dsh-ziXr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "This notebook demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your program. It covers:\n", - "\n", - "* Creating a `Dataset`.\n", - "* Iteration over a `Dataset` with eager execution enabled.\n", - "\n", - "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", - "\n", - "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", - "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", - "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." - ] - }, - { - "metadata": { - "id": "RlIWhyeLoYnG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "# Import TensorFlow.\n", - "import tensorflow as tf\n", - "\n", - "# Enable eager execution\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "H9UySOPLXdaw", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Create a source `Dataset`\n", - "\n", - "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." - ] - }, - { - "metadata": { - "id": "WPTUfGq6kJ5w", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", - "\n", - "# Create a CSV file\n", - "import tempfile\n", - "_, filename = tempfile.mkstemp()\n", - "with open(filename, 'w') as f:\n", - " f.write(\"\"\"Line 1\n", - "Line 2\n", - "Line 3\n", - " \"\"\")\n", - "ds_file = tf.data.TextLineDataset(filename)\n" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "twBfWd5xyu_d", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Apply transformations\n", - "\n", - "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." - ] - }, - { - "metadata": { - "id": "ngUe237Wt48W", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", - "ds_file = ds_file.batch(2)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "IDY4WsYRhP81", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Iterate\n", - "\n", - "When eager execution is enabled `Dataset` objects support iteration.\n", - "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." - ] - }, - { - "metadata": { - "id": "lCUWzso6mbqR", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "print('Elements of ds_tensors:')\n", - "for x in ds_tensors:\n", - " print(x)\n", - "\n", - "print('\\nElements in ds_file:')\n", - "for x in ds_file:\n", - " print(x)" - ], - "execution_count": 0, - "outputs": [] - } - ] -} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb index a00893bc77..5311a33e6e 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb @@ -3,7 +3,7 @@ "nbformat_minor": 0, "metadata": { "colab": { - "name": "Eager execution introduction", + "name": "eager_intro.ipynb", "version": "0.3.2", "views": {}, "default_view": {}, @@ -92,7 +92,8 @@ "\n", "* Importing required packages\n", "* Creating and using Tensors\n", - "* Using GPU acceleration" + "* Using GPU acceleration\n", + "* Datasets" ] }, { @@ -356,16 +357,135 @@ }, { "metadata": { - "id": "YEOJTNiOvnpQ", + "id": "o1K4dlhhHtQj", "colab_type": "text" }, "cell_type": "markdown", "source": [ - "## Next Steps\n", + "## Datasets\n", "\n", - "In this tutorial we covered the most fundamental concepts in TensorFlow - `Tensor`s, operations, and devices.\n", - "In [the next tutorial](https://github.com/tensorflow/models/tree/master/official/contrib/eager/python/examples/notebooks/2_gradients.ipynb) we will cover automatic differentiation - a building block required for training many machine learning models like neural networks." + "This section demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your model. It covers:\n", + "\n", + "* Creating a `Dataset`.\n", + "* Iteration over a `Dataset` with eager execution enabled.\n", + "\n", + "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", + "\n", + "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", + "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", + "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." + ] + }, + { + "metadata": { + "id": "zI0fmOynH-Ne", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Create a source `Dataset`\n", + "\n", + "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." + ] + }, + { + "metadata": { + "id": "F04fVOHQIBiG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", + "\n", + "# Create a CSV file\n", + "import tempfile\n", + "_, filename = tempfile.mkstemp()\n", + "\n", + "with open(filename, 'w') as f:\n", + " f.write(\"\"\"Line 1\n", + "Line 2\n", + "Line 3\n", + " \"\"\")\n", + "\n", + "ds_file = tf.data.TextLineDataset(filename)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "vbxIhC-5IPdf", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Apply transformations\n", + "\n", + "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." ] + }, + { + "metadata": { + "id": "uXSDZWE-ISsd", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", + "\n", + "ds_file = ds_file.batch(2)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "A8X1GNfoIZKJ", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Iterate\n", + "\n", + "When eager execution is enabled `Dataset` objects support iteration.\n", + "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." + ] + }, + { + "metadata": { + "id": "ws-WKRk5Ic6-", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "print('Elements of ds_tensors:')\n", + "for x in ds_tensors:\n", + " print(x)\n", + "\n", + "print('\\nElements in ds_file:')\n", + "for x in ds_file:\n", + " print(x)" + ], + "execution_count": 0, + "outputs": [] } ] } \ No newline at end of file -- cgit v1.2.3 From 280e40798a53a4296a990c0bc42edf8cfd8ac48b Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Thu, 28 Jun 2018 23:47:14 -0700 Subject: eager_intro title --- tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb index 5311a33e6e..a89b0e2c58 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb @@ -64,7 +64,7 @@ }, "cell_type": "markdown", "source": [ - "# Eager execution introduction" + "# Eager execution" ] }, { -- cgit v1.2.3 From 1c685650661690e70715ce214e9e23856ab463e7 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Fri, 29 Jun 2018 14:36:07 -0700 Subject: Reset get_started/_index --- tensorflow/docs_src/get_started/_index.yaml | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/tensorflow/docs_src/get_started/_index.yaml b/tensorflow/docs_src/get_started/_index.yaml index c3e216b2a6..4060804892 100644 --- a/tensorflow/docs_src/get_started/_index.yaml +++ b/tensorflow/docs_src/get_started/_index.yaml @@ -124,30 +124,30 @@ landing_page:
    1. {% dynamic if request.tld == 'cn' %} - Eager execution basics + Eager execution basics {% dynamic else %} - Eager execution basics + Eager execution basics {% dynamic endif %}
    2. {% dynamic if request.tld == 'cn' %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic else %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic endif %}
    3. {% dynamic if request.tld == 'cn' %} - Custom training: variables and models + Variables, models, and training {% dynamic else %} - Custom training: variables and models + Variables, models, and training {% dynamic endif %}
    4. {% dynamic if request.tld == 'cn' %} - Custom layers + Custom layers {% dynamic else %} - Custom layers + Custom layers {% dynamic endif %}
    5. Custom training walkthrough
      6. -- cgit v1.2.3 From 59f3c40a0729acee737f8dd9c4f8115ed42bb0f4 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Sat, 30 Jun 2018 10:44:39 -0700 Subject: Revert last commit that changed file names back. Now uses the correct path. This reverts commit 28c511013f960fd94cd0899342084570937ad277. --- tensorflow/docs_src/get_started/_index.yaml | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/tensorflow/docs_src/get_started/_index.yaml b/tensorflow/docs_src/get_started/_index.yaml index 4060804892..c3e216b2a6 100644 --- a/tensorflow/docs_src/get_started/_index.yaml +++ b/tensorflow/docs_src/get_started/_index.yaml @@ -124,30 +124,30 @@ landing_page:
      1. {% dynamic if request.tld == 'cn' %} - Eager execution basics + Eager execution basics {% dynamic else %} - Eager execution basics + Eager execution basics {% dynamic endif %}
      2. {% dynamic if request.tld == 'cn' %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic else %} - Automatic differentiation and gradient tapes + Automatic differentiation and gradient tapes {% dynamic endif %}
      3. {% dynamic if request.tld == 'cn' %} - Variables, models, and training + Custom training: variables and models {% dynamic else %} - Variables, models, and training + Custom training: variables and models {% dynamic endif %}
      4. {% dynamic if request.tld == 'cn' %} - Custom layers + Custom layers {% dynamic else %} - Custom layers + Custom layers {% dynamic endif %}
      5. Custom training walkthrough
        6. -- cgit v1.2.3 From ba18a05b6a5b4a695fe8dbd1b8bf564dac4e98a1 Mon Sep 17 00:00:00 2001 From: Billy Lamberta Date: Tue, 3 Jul 2018 10:41:41 -0700 Subject: Fix filename: eager_basics --- .../eager/python/examples/notebooks/README.md | 2 +- .../python/examples/notebooks/eager_basics.ipynb | 491 +++++++++++++++++++++ .../python/examples/notebooks/eager_intro.ipynb | 491 --------------------- 3 files changed, 492 insertions(+), 492 deletions(-) create mode 100644 tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb delete mode 100644 tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb diff --git a/tensorflow/contrib/eager/python/examples/notebooks/README.md b/tensorflow/contrib/eager/python/examples/notebooks/README.md index c17048d6ea..0d5ed84894 100644 --- a/tensorflow/contrib/eager/python/examples/notebooks/README.md +++ b/tensorflow/contrib/eager/python/examples/notebooks/README.md @@ -5,7 +5,7 @@ operations. Write custom layers, forward passes, and training loops with auto differentiation. Start with these notebooks, then read the [eager execution guide](https://www.tensorflow.org/guide/eager). -1. [Eager execution introduction](./eager_intro.ipynb) +1. [Eager execution basics](./eager_basics.ipynb) 2. [Automatic differentiation and gradient tapes](./automatic_differentiation.ipynb) 3. [Custom training: basics](./custom_training.ipynb) 4. [Custom layers](./custom_layers.ipynb) diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb new file mode 100644 index 0000000000..b37a18c9a6 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/notebooks/eager_basics.ipynb @@ -0,0 +1,491 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "eager_basics.ipynb", + "version": "0.3.2", + "views": {}, + "default_view": {}, + "provenance": [], + "private_outputs": true, + "collapsed_sections": [], + "toc_visible": true + }, + "kernelspec": { + "name": "python3", + "display_name": "Python 3" + } + }, + "cells": [ + { + "metadata": { + "id": "iPpI7RaYoZuE", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "##### Copyright 2018 The TensorFlow Authors." + ] + }, + { + "metadata": { + "id": "hro2InpHobKk", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "form" + }, + "cell_type": "code", + "source": [ + "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", + "# you may not use this file except in compliance with the License.\n", + "# You may obtain a copy of the License at\n", + "#\n", + "# https://www.apache.org/licenses/LICENSE-2.0\n", + "#\n", + "# Unless required by applicable law or agreed to in writing, software\n", + "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", + "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", + "# See the License for the specific language governing permissions and\n", + "# limitations under the License." + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "U9i2Dsh-ziXr", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Eager execution basics" + ] + }, + { + "metadata": { + "id": "Hndw-YcxoOJK", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "
        \n", + "\n", + " Run in Google Colab\n", + "\n", + "View source on GitHub
        " + ] + }, + { + "metadata": { + "id": "6sILUVbHoSgH", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "This is an introductory tutorial for using TensorFlow. It will cover:\n", + "\n", + "* Importing required packages\n", + "* Creating and using Tensors\n", + "* Using GPU acceleration\n", + "* Datasets" + ] + }, + { + "metadata": { + "id": "z1JcS5iBXMRO", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Import TensorFlow\n", + "\n", + "To get started, import the `tensorflow` module and enable eager execution.\n", + "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." + ] + }, + { + "metadata": { + "id": "RlIWhyeLoYnG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "\n", + "tf.enable_eager_execution()" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "H9UySOPLXdaw", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Tensors\n", + "\n", + "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" + ] + }, + { + "metadata": { + "id": "ngUe237Wt48W", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "print(tf.add(1, 2))\n", + "print(tf.add([1, 2], [3, 4]))\n", + "print(tf.square(5))\n", + "print(tf.reduce_sum([1, 2, 3]))\n", + "print(tf.encode_base64(\"hello world\"))\n", + "\n", + "# Operator overloading is also supported\n", + "print(tf.square(2) + tf.square(3))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "IDY4WsYRhP81", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "Each Tensor has a shape and a datatype" + ] + }, + { + "metadata": { + "id": "srYWH1MdJNG7", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "x = tf.matmul([[1]], [[2, 3]])\n", + "print(x.shape)\n", + "print(x.dtype)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "eBPw8e8vrsom", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", + "\n", + "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", + "2. Tensors are immutable." + ] + }, + { + "metadata": { + "id": "Dwi1tdW3JBw6", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### NumPy Compatibility\n", + "\n", + "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", + "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", + "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", + "\n", + "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", + "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." + ] + }, + { + "metadata": { + "id": "lCUWzso6mbqR", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "import numpy as np\n", + "\n", + "ndarray = np.ones([3, 3])\n", + "\n", + "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", + "tensor = tf.multiply(ndarray, 42)\n", + "print(tensor)\n", + "\n", + "\n", + "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", + "print(np.add(tensor, 1))\n", + "\n", + "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", + "print(tensor.numpy())" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "PBNP8yTRfu_X", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## GPU acceleration\n", + "\n", + "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" + ] + }, + { + "metadata": { + "id": "3Twf_Rw-gQFM", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + }, + "cellView": "code" + }, + "cell_type": "code", + "source": [ + "x = tf.random_uniform([3, 3])\n", + "\n", + "print(\"Is there a GPU available: \"),\n", + "print(tf.test.is_gpu_available())\n", + "\n", + "print(\"Is the Tensor on GPU #0: \"),\n", + "print(x.device.endswith('GPU:0'))" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "vpgYzgVXW2Ud", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Device Names\n", + "\n", + "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." + ] + }, + { + "metadata": { + "id": "ZWZQCimzuqyP", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "\n", + "\n", + "### Explicit Device Placement\n", + "\n", + "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" + ] + }, + { + "metadata": { + "id": "RjkNZTuauy-Q", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "def time_matmul(x):\n", + " %timeit tf.matmul(x, x)\n", + "\n", + "# Force execution on CPU\n", + "print(\"On CPU:\")\n", + "with tf.device(\"CPU:0\"):\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"CPU:0\")\n", + " time_matmul(x)\n", + "\n", + "# Force execution on GPU #0 if available\n", + "if tf.test.is_gpu_available():\n", + " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", + " x = tf.random_uniform([1000, 1000])\n", + " assert x.device.endswith(\"GPU:0\")\n", + " time_matmul(x)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "o1K4dlhhHtQj", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "## Datasets\n", + "\n", + "This section demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your model. It covers:\n", + "\n", + "* Creating a `Dataset`.\n", + "* Iteration over a `Dataset` with eager execution enabled.\n", + "\n", + "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", + "\n", + "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", + "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", + "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." + ] + }, + { + "metadata": { + "id": "zI0fmOynH-Ne", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Create a source `Dataset`\n", + "\n", + "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." + ] + }, + { + "metadata": { + "id": "F04fVOHQIBiG", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", + "\n", + "# Create a CSV file\n", + "import tempfile\n", + "_, filename = tempfile.mkstemp()\n", + "\n", + "with open(filename, 'w') as f:\n", + " f.write(\"\"\"Line 1\n", + "Line 2\n", + "Line 3\n", + " \"\"\")\n", + "\n", + "ds_file = tf.data.TextLineDataset(filename)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "vbxIhC-5IPdf", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Apply transformations\n", + "\n", + "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." + ] + }, + { + "metadata": { + "id": "uXSDZWE-ISsd", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", + "\n", + "ds_file = ds_file.batch(2)" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "A8X1GNfoIZKJ", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "### Iterate\n", + "\n", + "When eager execution is enabled `Dataset` objects support iteration.\n", + "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." + ] + }, + { + "metadata": { + "id": "ws-WKRk5Ic6-", + "colab_type": "code", + "colab": { + "autoexec": { + "startup": false, + "wait_interval": 0 + } + } + }, + "cell_type": "code", + "source": [ + "print('Elements of ds_tensors:')\n", + "for x in ds_tensors:\n", + " print(x)\n", + "\n", + "print('\\nElements in ds_file:')\n", + "for x in ds_file:\n", + " print(x)" + ], + "execution_count": 0, + "outputs": [] + } + ] +} \ No newline at end of file diff --git a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb b/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb deleted file mode 100644 index a89b0e2c58..0000000000 --- a/tensorflow/contrib/eager/python/examples/notebooks/eager_intro.ipynb +++ /dev/null @@ -1,491 +0,0 @@ -{ - "nbformat": 4, - "nbformat_minor": 0, - "metadata": { - "colab": { - "name": "eager_intro.ipynb", - "version": "0.3.2", - "views": {}, - "default_view": {}, - "provenance": [], - "private_outputs": true, - "collapsed_sections": [], - "toc_visible": true - }, - "kernelspec": { - "name": "python3", - "display_name": "Python 3" - } - }, - "cells": [ - { - "metadata": { - "id": "iPpI7RaYoZuE", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "##### Copyright 2018 The TensorFlow Authors." - ] - }, - { - "metadata": { - "id": "hro2InpHobKk", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "form" - }, - "cell_type": "code", - "source": [ - "#@title Licensed under the Apache License, Version 2.0 (the \"License\");\n", - "# you may not use this file except in compliance with the License.\n", - "# You may obtain a copy of the License at\n", - "#\n", - "# https://www.apache.org/licenses/LICENSE-2.0\n", - "#\n", - "# Unless required by applicable law or agreed to in writing, software\n", - "# distributed under the License is distributed on an \"AS IS\" BASIS,\n", - "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n", - "# See the License for the specific language governing permissions and\n", - "# limitations under the License." - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "U9i2Dsh-ziXr", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "# Eager execution" - ] - }, - { - "metadata": { - "id": "Hndw-YcxoOJK", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "
        \n", - "\n", - " Run in Google Colab\n", - "\n", - "View source on GitHub
        " - ] - }, - { - "metadata": { - "id": "6sILUVbHoSgH", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "This is an introductory tutorial for using TensorFlow. It will cover:\n", - "\n", - "* Importing required packages\n", - "* Creating and using Tensors\n", - "* Using GPU acceleration\n", - "* Datasets" - ] - }, - { - "metadata": { - "id": "z1JcS5iBXMRO", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Import TensorFlow\n", - "\n", - "To get started, import the `tensorflow` module and enable eager execution.\n", - "Eager execution enables a more interactive frontend to TensorFlow, the details of which we will discuss much later." - ] - }, - { - "metadata": { - "id": "RlIWhyeLoYnG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "import tensorflow as tf\n", - "\n", - "tf.enable_eager_execution()" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "H9UySOPLXdaw", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Tensors\n", - "\n", - "A Tensor is a multi-dimensional array. Similar to NumPy `ndarray` objects, `Tensor` objects have a data type and a shape. Additionally, Tensors can reside in accelerator (like GPU) memory. TensorFlow offers a rich library of operations ([tf.add](https://www.tensorflow.org/api_docs/python/tf/add), [tf.matmul](https://www.tensorflow.org/api_docs/python/tf/matmul), [tf.linalg.inv](https://www.tensorflow.org/api_docs/python/tf/linalg/inv) etc.) that consume and produce Tensors. These operations automatically convert native Python types. For example:\n" - ] - }, - { - "metadata": { - "id": "ngUe237Wt48W", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "print(tf.add(1, 2))\n", - "print(tf.add([1, 2], [3, 4]))\n", - "print(tf.square(5))\n", - "print(tf.reduce_sum([1, 2, 3]))\n", - "print(tf.encode_base64(\"hello world\"))\n", - "\n", - "# Operator overloading is also supported\n", - "print(tf.square(2) + tf.square(3))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "IDY4WsYRhP81", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "Each Tensor has a shape and a datatype" - ] - }, - { - "metadata": { - "id": "srYWH1MdJNG7", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "x = tf.matmul([[1]], [[2, 3]])\n", - "print(x.shape)\n", - "print(x.dtype)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "eBPw8e8vrsom", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "The most obvious differences between NumPy arrays and TensorFlow Tensors are:\n", - "\n", - "1. Tensors can be backed by accelerator memory (like GPU, TPU).\n", - "2. Tensors are immutable." - ] - }, - { - "metadata": { - "id": "Dwi1tdW3JBw6", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### NumPy Compatibility\n", - "\n", - "Conversion between TensorFlow Tensors and NumPy ndarrays is quite simple as:\n", - "* TensorFlow operations automatically convert NumPy ndarrays to Tensors.\n", - "* NumPy operations automatically convert Tensors to NumPy ndarrays.\n", - "\n", - "Tensors can be explicitly converted to NumPy ndarrays by invoking the `.numpy()` method on them.\n", - "These conversions are typically cheap as the array and Tensor share the underlying memory representation if possible. However, sharing the underlying representation isn't always possible since the Tensor may be hosted in GPU memory while NumPy arrays are always backed by host memory, and the conversion will thus involve a copy from GPU to host memory." - ] - }, - { - "metadata": { - "id": "lCUWzso6mbqR", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "import numpy as np\n", - "\n", - "ndarray = np.ones([3, 3])\n", - "\n", - "print(\"TensorFlow operations convert numpy arrays to Tensors automatically\")\n", - "tensor = tf.multiply(ndarray, 42)\n", - "print(tensor)\n", - "\n", - "\n", - "print(\"And NumPy operations convert Tensors to numpy arrays automatically\")\n", - "print(np.add(tensor, 1))\n", - "\n", - "print(\"The .numpy() method explicitly converts a Tensor to a numpy array\")\n", - "print(tensor.numpy())" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "PBNP8yTRfu_X", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## GPU acceleration\n", - "\n", - "Many TensorFlow operations can be accelerated by using the GPU for computation. Without any annotations, TensorFlow automatically decides whether to use the GPU or CPU for an operation (and copies the tensor between CPU and GPU memory if necessary). Tensors produced by an operation are typically backed by the memory of the device on which the operation executed. For example:" - ] - }, - { - "metadata": { - "id": "3Twf_Rw-gQFM", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - }, - "cellView": "code" - }, - "cell_type": "code", - "source": [ - "x = tf.random_uniform([3, 3])\n", - "\n", - "print(\"Is there a GPU available: \"),\n", - "print(tf.test.is_gpu_available())\n", - "\n", - "print(\"Is the Tensor on GPU #0: \"),\n", - "print(x.device.endswith('GPU:0'))" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "vpgYzgVXW2Ud", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Device Names\n", - "\n", - "The `Tensor.device` property provides a fully qualified string name of the device hosting the contents of the Tensor. This name encodes a bunch of details, such as an identifier of the network address of the host on which this program is executing and the device within that host. This is required for distributed execution of TensorFlow programs, but we'll skip that for now. The string will end with `GPU:` if the tensor is placed on the `N`-th tensor on the host." - ] - }, - { - "metadata": { - "id": "ZWZQCimzuqyP", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "\n", - "\n", - "### Explicit Device Placement\n", - "\n", - "The term \"placement\" in TensorFlow refers to how individual operations are assigned (placed on) a device for execution. As mentioned above, when there is no explicit guidance provided, TensorFlow automatically decides which device to execute an operation, and copies Tensors to that device if needed. However, TensorFlow operations can be explicitly placed on specific devices using the `tf.device` context manager. For example:" - ] - }, - { - "metadata": { - "id": "RjkNZTuauy-Q", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "def time_matmul(x):\n", - " %timeit tf.matmul(x, x)\n", - "\n", - "# Force execution on CPU\n", - "print(\"On CPU:\")\n", - "with tf.device(\"CPU:0\"):\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"CPU:0\")\n", - " time_matmul(x)\n", - "\n", - "# Force execution on GPU #0 if available\n", - "if tf.test.is_gpu_available():\n", - " with tf.device(\"GPU:0\"): # Or GPU:1 for the 2nd GPU, GPU:2 for the 3rd etc.\n", - " x = tf.random_uniform([1000, 1000])\n", - " assert x.device.endswith(\"GPU:0\")\n", - " time_matmul(x)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "o1K4dlhhHtQj", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "## Datasets\n", - "\n", - "This section demonstrates the use of the [`tf.data.Dataset` API](https://www.tensorflow.org/guide/datasets) to build pipelines to feed data to your model. It covers:\n", - "\n", - "* Creating a `Dataset`.\n", - "* Iteration over a `Dataset` with eager execution enabled.\n", - "\n", - "We recommend using the `Dataset`s API for building performant, complex input pipelines from simple, re-usable pieces that will feed your model's training or evaluation loops.\n", - "\n", - "If you're familiar with TensorFlow graphs, the API for constructing the `Dataset` object remains exactly the same when eager execution is enabled, but the process of iterating over elements of the dataset is slightly simpler.\n", - "You can use Python iteration over the `tf.data.Dataset` object and do not need to explicitly create an `tf.data.Iterator` object.\n", - "As a result, the discussion on iterators in the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets) is not relevant when eager execution is enabled." - ] - }, - { - "metadata": { - "id": "zI0fmOynH-Ne", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Create a source `Dataset`\n", - "\n", - "Create a _source_ dataset using one of the factory functions like [`Dataset.from_tensors`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensors), [`Dataset.from_tensor_slices`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_tensor_slices) or using objects that read from files like [`TextLineDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TextLineDataset) or [`TFRecordDataset`](https://www.tensorflow.org/api_docs/python/tf/data/TFRecordDataset). See the [TensorFlow Guide](https://www.tensorflow.org/guide/datasets#reading_input_data) for more information." - ] - }, - { - "metadata": { - "id": "F04fVOHQIBiG", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "ds_tensors = tf.data.Dataset.from_tensor_slices([1, 2, 3, 4, 5, 6])\n", - "\n", - "# Create a CSV file\n", - "import tempfile\n", - "_, filename = tempfile.mkstemp()\n", - "\n", - "with open(filename, 'w') as f:\n", - " f.write(\"\"\"Line 1\n", - "Line 2\n", - "Line 3\n", - " \"\"\")\n", - "\n", - "ds_file = tf.data.TextLineDataset(filename)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "vbxIhC-5IPdf", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Apply transformations\n", - "\n", - "Use the transformations functions like [`map`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#map), [`batch`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#batch), [`shuffle`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset#shuffle) etc. to apply transformations to the records of the dataset. See the [API documentation for `tf.data.Dataset`](https://www.tensorflow.org/api_docs/python/tf/data/Dataset) for details." - ] - }, - { - "metadata": { - "id": "uXSDZWE-ISsd", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "ds_tensors = ds_tensors.map(tf.square).shuffle(2).batch(2)\n", - "\n", - "ds_file = ds_file.batch(2)" - ], - "execution_count": 0, - "outputs": [] - }, - { - "metadata": { - "id": "A8X1GNfoIZKJ", - "colab_type": "text" - }, - "cell_type": "markdown", - "source": [ - "### Iterate\n", - "\n", - "When eager execution is enabled `Dataset` objects support iteration.\n", - "If you're familiar with the use of `Dataset`s in TensorFlow graphs, note that there is no need for calls to `Dataset.make_one_shot_iterator()` or `get_next()` calls." - ] - }, - { - "metadata": { - "id": "ws-WKRk5Ic6-", - "colab_type": "code", - "colab": { - "autoexec": { - "startup": false, - "wait_interval": 0 - } - } - }, - "cell_type": "code", - "source": [ - "print('Elements of ds_tensors:')\n", - "for x in ds_tensors:\n", - " print(x)\n", - "\n", - "print('\\nElements in ds_file:')\n", - "for x in ds_file:\n", - " print(x)" - ], - "execution_count": 0, - "outputs": [] - } - ] -} \ No newline at end of file -- cgit v1.2.3