diff options
Diffstat (limited to 'tensorflow/contrib/eager/python/examples/workshop/1_basic.ipynb')
-rw-r--r-- | tensorflow/contrib/eager/python/examples/workshop/1_basic.ipynb | 282 |
1 files changed, 282 insertions, 0 deletions
diff --git a/tensorflow/contrib/eager/python/examples/workshop/1_basic.ipynb b/tensorflow/contrib/eager/python/examples/workshop/1_basic.ipynb new file mode 100644 index 0000000000..75cb3f8227 --- /dev/null +++ b/tensorflow/contrib/eager/python/examples/workshop/1_basic.ipynb @@ -0,0 +1,282 @@ +{ + "nbformat": 4, + "nbformat_minor": 0, + "metadata": { + "colab": { + "name": "TFE Workshop: control flow", + "version": "0.3.2", + "provenance": [], + "include_colab_link": true + } + }, + "cells": [ + { + "cell_type": "markdown", + "metadata": { + "id": "view-in-github", + "colab_type": "text" + }, + "source": [ + "[View in Colaboratory](https://colab.research.google.com/gist/alextp/664b2f8700485ff6801f4d26293bd567/tfe-workshop-control-flow.ipynb)" + ] + }, + { + "metadata": { + "id": "9BpQzh9BvJlj", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 37 + }, + "outputId": "0b336886-8204-4815-89fa-5291a49d5784" + }, + "cell_type": "code", + "source": [ + "import tensorflow as tf\n", + "import numpy as np\n", + "tf.enable_eager_execution()" + ], + "execution_count": 1, + "outputs": [] + }, + { + "metadata": { + "id": "0roIB19GvOjI", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Eager execution basics\n", + "\n", + "When eager execution is enabled TensorFlow immediately executes operations, and Tensors are always available. " + ] + }, + { + "metadata": { + "id": "jeO8F-V-vN24", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 68 + }, + "outputId": "aeb3bdec-50b7-440d-93d8-5a171f091081" + }, + "cell_type": "code", + "source": [ + "t = tf.constant([[1, 2], [3, 4]])\n", + "t" + ], + "execution_count": 2, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "<tf.Tensor: id=0, shape=(2, 2), dtype=int32, numpy=\n", + "array([[1, 2],\n", + " [3, 4]], dtype=int32)>" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 2 + } + ] + }, + { + "metadata": { + "id": "Y17RwSFxvlDL", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 68 + }, + "outputId": "cfcc10c7-707b-4997-99b3-a5f382c5166b" + }, + "cell_type": "code", + "source": [ + "tf.matmul(t, t)" + ], + "execution_count": 3, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "<tf.Tensor: id=2, shape=(2, 2), dtype=int32, numpy=\n", + "array([[ 7, 10],\n", + " [15, 22]], dtype=int32)>" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 3 + } + ] + }, + { + "metadata": { + "id": "Dab1bS3TvmRE", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 34 + }, + "outputId": "8a624f3d-a658-4359-c586-1c5f6bf4c8b7" + }, + "cell_type": "code", + "source": [ + "# It's also possible to have Python control flow which depends on the value of tensors.\n", + "if t[0, 0] > 0.5:\n", + " print(\"T is bigger\")\n", + "else:\n", + " print(\"T is smaller\")" + ], + "execution_count": 4, + "outputs": [ + { + "output_type": "stream", + "text": [ + "T is bigger\n" + ], + "name": "stdout" + } + ] + }, + { + "metadata": { + "id": "dPgptJcGwIon", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 34 + }, + "outputId": "c4f27f2b-0848-4475-dde5-2534dac65a5c" + }, + "cell_type": "code", + "source": [ + "# Tensors are also usable as numpy arrays\n", + "np.prod(t)" + ], + "execution_count": 6, + "outputs": [ + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "24" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 6 + } + ] + }, + { + "metadata": { + "id": "p3DTfQXnwXzj", + "colab_type": "text" + }, + "cell_type": "markdown", + "source": [ + "# Exercise\n", + "\n", + "The algorithm for bisecting line search is a pretty simple way to find a zero of a continuous scalar function in an interval [a,b] where f(a) and f(b) have different signs. Simply evaluate f((a+b)/2), and narrow the interval by replacing either a or b with (a+b)/2 such that the function when applied on the boundary of the interval still has different signs.\n", + "\n", + "Implement a python function `bisecting_line_search(f, a, b, epsilon)` which returns a value such that `tf.abs(f(value)) < epsilon`.\n", + "\n", + "One thing to keep in mind: python's `==` opertor is not overloaded on Tensors, so you need to use `tf.equal` to compare for equality." + ] + }, + { + "metadata": { + "id": "6eq0YuI6ykm5", + "colab_type": "code", + "colab": {} + }, + "cell_type": "code", + "source": [ + "# Example test harness to get you going\n", + "\n", + "def test_f(x):\n", + " return x - 0.1234\n", + "def bisecting_line_search(f, a, b, epsilon):\n", + " # Return x such that f(x) <= epsilon.\n", + " pass\n", + "a = tf.constant(0.0)\n", + "b = tf.constant(1.0)\n", + "epsilon = tf.constant(0.001)\n", + "x = bisecting_line_search(test_f, a, b, epsilon)\n" + ], + "execution_count": 0, + "outputs": [] + }, + { + "metadata": { + "id": "LcMmEfd_xvej", + "colab_type": "code", + "colab": { + "base_uri": "https://localhost:8080/", + "height": 170 + }, + "outputId": "f402aa50-8ce3-4416-f755-8bbcd1af7809" + }, + "cell_type": "code", + "source": [ + "#@title Double-click to see the solution\n", + "\n", + "def bisecting_line_search(f, a, b, epsilon):\n", + " f_a = f(a)\n", + " f_b = f(b)\n", + " probe = (a + b) / 2\n", + " f_probe = f(probe)\n", + " while tf.abs(f_probe) > epsilon:\n", + " if tf.equal(tf.sign(f_probe), tf.sign(f_a)):\n", + " a = probe\n", + " f_a = f_probe\n", + " else:\n", + " b = probe\n", + " f_b = f_probe\n", + " probe = (a + b) / 2\n", + " f_probe = f(probe)\n", + " print(\"new probe\", probe)\n", + " return probe\n", + "\n", + "bisecting_line_search(test_f, 0., 1., 0.001)" + ], + "execution_count": 8, + "outputs": [ + { + "output_type": "stream", + "text": [ + "('new probe', 0.25)\n", + "('new probe', 0.125)\n", + "('new probe', 0.0625)\n", + "('new probe', 0.09375)\n", + "('new probe', 0.109375)\n", + "('new probe', 0.1171875)\n", + "('new probe', 0.12109375)\n", + "('new probe', 0.123046875)\n" + ], + "name": "stdout" + }, + { + "output_type": "execute_result", + "data": { + "text/plain": [ + "0.123046875" + ] + }, + "metadata": { + "tags": [] + }, + "execution_count": 8 + } + ] + } + ] +} |