path: root/tensorflow/contrib/eager/python/examples/workshop/2_models.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "TFE Workshop: Models.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "[View in Colaboratory](https://colab.research.google.com/gist/alextp/5cfcffd408bd5103f5ae747bc97ab0b5/tfe-workshop-models.ipynb)"
      ]
    },
    {
      "metadata": {
        "id": "BMxv1O6Q0SJL",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        },
        "outputId": "8be9c556-ac7f-4142-e35e-19dc2b097121"
      },
      "cell_type": "code",
      "source": [
        "import tensorflow as tf\n",
        "tf.enable_eager_execution()\n",
        "tfe = tf.contrib.eager"
      ],
      "execution_count": 1,
      "outputs": []
    },
    {
      "metadata": {
        "id": "lE1vJhxp0WR9",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Variables\n",
        "\n",
        "TensorFlow variables are useful to store the state in your program. They are integrated with other parts of the API (taking gradients, checkpointing, graph functions)."
      ]
    },
    {
      "metadata": {
        "id": "C4ztQNgc0VpW",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "8b63ae1f-2670-49c0-a31b-8cf7fc4194a1"
      },
      "cell_type": "code",
      "source": [
        "# Creating variables\n",
        "v = tf.Variable(1.0)\n",
        "v"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Variable 'Variable:0' shape=() dtype=float32, numpy=1.0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        }
      ]
    },
    {
      "metadata": {
        "id": "H0daItGg1IAp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "e47d5aab-16a1-4e29-c27d-7fbc0b94b5d3"
      },
      "cell_type": "code",
      "source": [
        "v.assign_add(1.0)\n",
        "v"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Variable 'Variable:0' shape=() dtype=float32, numpy=2.0>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "metadata": {
        "id": "BJvBzcIG1hyK",
        "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": "iSQTS3QW1YQQ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        },
        "outputId": "c5d8aa10-dcad-44f7-f0eb-0faf5249fd7e"
      },
      "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",
        "\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))\n"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "metadata": {
        "id": "nRuUogoS1liV",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        },
        "outputId": "c352ce79-d519-45e4-a12e-1eaba76871a2"
      },
      "cell_type": "code",
      "source": [
        "layer(tf.zeros([2, 2]))"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor: id=43, shape=(2, 10), dtype=float32, numpy=\n",
              "array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
              "       [0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]], dtype=float32)>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "metadata": {
        "id": "JH4Kf4ka1mht",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 136
        },
        "outputId": "c34e2378-f83d-42c5-d30a-ebe55620368a"
      },
      "cell_type": "code",
      "source": [
        "layer.variables"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<tf.Variable 'dense/kernel:0' shape=(2, 10) dtype=float32, numpy=\n",
              " array([[-0.42494273, -0.2067694 ,  0.4519381 ,  0.6842533 ,  0.04131705,\n",
              "          0.70547956,  0.4021917 , -0.5939298 , -0.5671462 ,  0.5586321 ],\n",
              "        [ 0.3709975 , -0.64126074, -0.5386696 , -0.42212513,  0.6550072 ,\n",
              "          0.70081085,  0.08859557, -0.30801034, -0.31450653,  0.02522504]],\n",
              "       dtype=float32)>,\n",
              " <tf.Variable 'dense/bias:0' shape=(10,) dtype=float32, numpy=array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "metadata": {
        "id": "DSI4NF0_1vn-",
        "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": "hMgDBftJ12Bp",
        "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.\n"
      ]
    },
    {
      "metadata": {
        "id": "K3gVY6gj1nbe",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 190
        },
        "outputId": "6e9be0c4-960e-46c2-cdd9-7e94ad09d46b"
      },
      "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",
        "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": 7,
      "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/conv2d/kernel:0', 'resnet_identity_block/conv2d/bias:0', 'resnet_identity_block/batch_normalization/gamma:0', 'resnet_identity_block/batch_normalization/beta:0', 'resnet_identity_block/conv2d_1/kernel:0', 'resnet_identity_block/conv2d_1/bias:0', 'resnet_identity_block/batch_normalization_1/gamma:0', 'resnet_identity_block/batch_normalization_1/beta:0', 'resnet_identity_block/conv2d_2/kernel:0', 'resnet_identity_block/conv2d_2/bias:0', 'resnet_identity_block/batch_normalization_2/gamma:0', 'resnet_identity_block/batch_normalization_2/beta:0', 'resnet_identity_block/batch_normalization/moving_mean:0', 'resnet_identity_block/batch_normalization/moving_variance:0', 'resnet_identity_block/batch_normalization_1/moving_mean:0', 'resnet_identity_block/batch_normalization_1/moving_variance:0', 'resnet_identity_block/batch_normalization_2/moving_mean:0', 'resnet_identity_block/batch_normalization_2/moving_variance:0']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "LPXhHUIc1-sO",
        "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": "5pXgzNAU17xk",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 173
        },
        "outputId": "03b7eaf8-9b35-482b-bcf0-a99af6c2c6a4"
      },
      "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]))\n"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor: id=493, shape=(1, 2, 3, 3), dtype=float32, numpy=\n",
              "array([[[[0., 0., 0.],\n",
              "         [0., 0., 0.],\n",
              "         [0., 0., 0.]],\n",
              "\n",
              "        [[0., 0., 0.],\n",
              "         [0., 0., 0.],\n",
              "         [0., 0., 0.]]]], dtype=float32)>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "metadata": {
        "id": "MZrns6p22GEQ",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Exercise!\n",
        "\n",
        "Make a simple convolutional neural network model, useful for things such as MNIST which don't need too many parameters. A sequence of two or three convolutions with small output channels (say, 32 and 64) plus one or two fully connected layers is probably enough.\n",
        "\n",
        "The input shape should be [batch_size, 28, 28, 1]."
      ]
    },
    {
      "metadata": {
        "id": "8CAUa3KNN916",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        },
        "outputId": "97c0ff3c-c962-4c13-eee8-406101465761"
      },
      "cell_type": "code",
      "source": [
        "# TODO: Implement a convolutional model as described above, and assign it to\n",
        "# model.\n",
        "model = tf.keras.Sequential([\n",
        "    \n",
        "])"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "metadata": {
        "id": "vLDDduR32E82",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "09bb1d43-b4c6-44b5-916e-0d2903d10cf4"
      },
      "cell_type": "code",
      "source": [
        "#@title Click to see the answer\n",
        "\n",
        "max_pool = tf.keras.layers.MaxPooling2D(\n",
        "      (2, 2), (2, 2), padding='same')\n",
        "  # The model consists of a sequential chain of layers, so tf.keras.Sequential\n",
        "  # (a subclass of tf.keras.Model) makes for a compact description.\n",
        "model = tf.keras.Sequential(\n",
        "      [\n",
        "          tf.keras.layers.Conv2D(\n",
        "              32,\n",
        "              5,\n",
        "              padding='same',\n",
        "              activation=tf.nn.relu),\n",
        "          max_pool,\n",
        "          tf.keras.layers.Conv2D(\n",
        "              64,\n",
        "              5,\n",
        "              padding='same',\n",
        "              activation=tf.nn.relu),\n",
        "          max_pool,\n",
        "          tf.keras.layers.Flatten(),\n",
        "          tf.keras.layers.Dense(1024, activation=tf.nn.relu),\n",
        "          tf.keras.layers.Dropout(0.4),\n",
        "          tf.keras.layers.Dense(10)\n",
        "      ])\n",
        "\n",
        "model(tf.zeros([1, 28, 28, 1]))"
      ],
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor: id=625, shape=(1, 10), dtype=float32, numpy=array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]], dtype=float32)>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 10
        }
      ]
    },
    {
      "metadata": {
        "id": "H_CKVBroik4M",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Stop here for now"
      ]
    },
    {
      "metadata": {
        "id": "_yRwuE6MMmzC",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Training\n",
        "\n",
        "When eager execution is enabled, you can write Pythonic training loops. Simply\n",
        "\n",
        "1. load your data into a `tf.data.Dataset`, which lets you construct functional pipelines for processing, shuffling, and batching your data,\n",
        "2. iterate over the dataset using a Python `for` loop, and\n",
        "3. perform an optimization step in the body of your `for` loop.\n",
        "\n",
        "This workflow is exemplified in the following exercise."
      ]
    },
    {
      "metadata": {
        "id": "gj0-EkTc_Xt1",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "\n",
        "\n",
        "## Exercise!\n",
        "\n",
        "In this exercise, you'll train the convolutional model you implemented for the previous exericse on the MNIST dataset. "
      ]
    },
    {
      "metadata": {
        "id": "WOGm9HHn_byR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 17
        },
        "outputId": "bbccc7ad-33cd-446e-bcda-f358c7547e1b"
      },
      "cell_type": "code",
      "source": [
        "#@title Utilities for downloading MNIST data (double-click to show code)\n",
        "import gzip\n",
        "import os\n",
        "import tempfile\n",
        "from six.moves import urllib\n",
        "import shutil\n",
        "\n",
        "import numpy as np\n",
        "\n",
        "def read32(bytestream):\n",
        "  \"\"\"Read 4 bytes from bytestream as an unsigned 32-bit integer.\"\"\"\n",
        "  dt = np.dtype(np.uint32).newbyteorder('>')\n",
        "  return np.frombuffer(bytestream.read(4), dtype=dt)[0]\n",
        "\n",
        "\n",
        "def check_image_file_header(filename):\n",
        "  \"\"\"Validate that filename corresponds to images for the MNIST dataset.\"\"\"\n",
        "  with tf.gfile.Open(filename, 'rb') as f:\n",
        "    magic = read32(f)\n",
        "    read32(f)  # num_images, unused\n",
        "    rows = read32(f)\n",
        "    cols = read32(f)\n",
        "    if magic != 2051:\n",
        "      raise ValueError('Invalid magic number %d in MNIST file %s' % (magic,\n",
        "                                                                     f.name))\n",
        "    if rows != 28 or cols != 28:\n",
        "      raise ValueError(\n",
        "          'Invalid MNIST file %s: Expected 28x28 images, found %dx%d' %\n",
        "          (f.name, rows, cols))\n",
        "\n",
        "\n",
        "def check_labels_file_header(filename):\n",
        "  \"\"\"Validate that filename corresponds to labels for the MNIST dataset.\"\"\"\n",
        "  with tf.gfile.Open(filename, 'rb') as f:\n",
        "    magic = read32(f)\n",
        "    read32(f)  # num_items, unused\n",
        "    if magic != 2049:\n",
        "      raise ValueError('Invalid magic number %d in MNIST file %s' % (magic,\n",
        "                                                                     f.name))\n",
        "      \n",
        "def download(directory, filename):\n",
        "  \"\"\"Download (and unzip) a file from the MNIST dataset if not already done.\"\"\"\n",
        "  filepath = os.path.join(directory, filename)\n",
        "  if tf.gfile.Exists(filepath):\n",
        "    return filepath\n",
        "  if not tf.gfile.Exists(directory):\n",
        "    tf.gfile.MakeDirs(directory)\n",
        "  # CVDF mirror of http://yann.lecun.com/exdb/mnist/\n",
        "  url = 'https://storage.googleapis.com/cvdf-datasets/mnist/' + filename + '.gz'\n",
        "  _, zipped_filepath = tempfile.mkstemp(suffix='.gz')\n",
        "  print('Downloading %s to %s' % (url, zipped_filepath))\n",
        "  urllib.request.urlretrieve(url, zipped_filepath)\n",
        "  with gzip.open(zipped_filepath, 'rb') as f_in, \\\n",
        "      tf.gfile.Open(filepath, 'wb') as f_out:\n",
        "    shutil.copyfileobj(f_in, f_out)\n",
        "  os.remove(zipped_filepath)\n",
        "  return filepath\n",
        "\n",
        "\n",
        "def dataset(directory, images_file, labels_file):\n",
        "  \"\"\"Download and parse MNIST dataset.\"\"\"\n",
        "\n",
        "  images_file = download(directory, images_file)\n",
        "  labels_file = download(directory, labels_file)\n",
        "\n",
        "  check_image_file_header(images_file)\n",
        "  check_labels_file_header(labels_file)\n",
        "\n",
        "  def decode_image(image):\n",
        "    # Normalize from [0, 255] to [0.0, 1.0]\n",
        "    image = tf.decode_raw(image, tf.uint8)\n",
        "    image = tf.cast(image, tf.float32)\n",
        "    image = tf.reshape(image, [28, 28, 1])\n",
        "    return image / 255.0\n",
        "\n",
        "  def decode_label(label):\n",
        "    label = tf.decode_raw(label, tf.uint8)  # tf.string -> [tf.uint8]\n",
        "    label = tf.reshape(label, [])  # label is a scalar\n",
        "    return tf.to_int32(label)\n",
        "\n",
        "  images = tf.data.FixedLengthRecordDataset(\n",
        "      images_file, 28 * 28, header_bytes=16).map(decode_image)\n",
        "  labels = tf.data.FixedLengthRecordDataset(\n",
        "      labels_file, 1, header_bytes=8).map(decode_label)\n",
        "  return tf.data.Dataset.zip((images, labels))\n",
        "\n",
        "\n",
        "def get_training_data(directory):\n",
        "  \"\"\"tf.data.Dataset object for MNIST training data.\"\"\"\n",
        "  return dataset(directory, 'train-images-idx3-ubyte',\n",
        "                 'train-labels-idx1-ubyte').take(1024)\n",
        "\n",
        "def get_test_data(directory):\n",
        "  \"\"\"tf.data.Dataset object for MNIST test data.\"\"\"\n",
        "  return dataset(directory, 't10k-images-idx3-ubyte', 't10k-labels-idx1-ubyte')"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "metadata": {
        "id": "4ejmJ2dv_f0R",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "274c0381-e505-4e69-f910-3def6f8572a7"
      },
      "cell_type": "code",
      "source": [
        "# Don't forget to run the cell above!\n",
        "training_data = get_training_data(\"/tmp/mnist/train\")\n",
        "test_data = get_test_data(\"/tmp/mnist/test\")"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading https://storage.googleapis.com/cvdf-datasets/mnist/train-images-idx3-ubyte.gz to /tmp/tmp4ull1xwa.gz\n",
            "Downloading https://storage.googleapis.com/cvdf-datasets/mnist/train-labels-idx1-ubyte.gz to /tmp/tmp1eikhj1v.gz\n",
            "Downloading https://storage.googleapis.com/cvdf-datasets/mnist/t10k-images-idx3-ubyte.gz to /tmp/tmpcp8xah9c.gz\n",
            "Downloading https://storage.googleapis.com/cvdf-datasets/mnist/t10k-labels-idx1-ubyte.gz to /tmp/tmpqww_1e74.gz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "TANpFS6GKLMC",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Fill in the implementation of `train_one_epoch` below and run the cell to train your model. "
      ]
    },
    {
      "metadata": {
        "id": "btKL0Ss9_rmC",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "56858516-86fc-424a-f00d-6f088f98bf9b"
      },
      "cell_type": "code",
      "source": [
        "EPOCHS = 5\n",
        "optimizer = tf.train.MomentumOptimizer(learning_rate=0.01, momentum=0.5)\n",
        "\n",
        "def loss_fn(logits, labels):\n",
        "  return tf.reduce_mean(\n",
        "      tf.nn.sparse_softmax_cross_entropy_with_logits(\n",
        "          logits=tf.squeeze(logits), labels=labels))\n",
        "\n",
        "def train_one_epoch(model, training_data, optimizer):\n",
        "  # TODO: Implement an optimization step and return the average loss.\n",
        "  #\n",
        "  # Hint: Use `tf.GradientTape` to compute the gradient of the loss, and use\n",
        "  # `optimizer.apply_gradients` to update the model's variables, which are\n",
        "  #  accessible as `model.variables`\n",
        "  average_loss = tfe.metrics.Mean('loss')\n",
        "  for images, labels in training_data.shuffle(buffer_size=10000).batch(64):\n",
        "    pass\n",
        "  return average_loss.result()\n",
        "\n",
        "for epoch in range(EPOCHS):\n",
        "  loss = train_one_epoch(model, training_data, optimizer)\n",
        "  print(\"Average loss after epoch %d: %.4f\" % (epoch, loss))"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Average loss after epoch 0: 2.2847\n",
            "Average loss after epoch 1: 2.2305\n",
            "Average loss after epoch 2: 2.1334\n",
            "Average loss after epoch 3: 1.9115\n",
            "Average loss after epoch 4: 1.4285\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "yAOFupJN_htg",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "67e711e4-76c9-4e3f-bb49-a14955dba03a"
      },
      "cell_type": "code",
      "source": [
        "#@title Double-click to see a solution.\n",
        "EPOCHS = 5\n",
        "optimizer = tf.train.MomentumOptimizer(learning_rate=0.01, momentum=0.5)\n",
        "\n",
        "def _loss_fn(logits, labels):\n",
        "  return tf.reduce_mean(\n",
        "      tf.nn.sparse_softmax_cross_entropy_with_logits(\n",
        "          logits=tf.squeeze(logits), labels=labels))\n",
        "\n",
        "def _train_one_epoch(model, training_data):\n",
        "  average_loss = tfe.metrics.Mean(\"loss\")\n",
        "  for images, labels in training_data.shuffle(buffer_size=10000).batch(64):\n",
        "    with tf.GradientTape() as tape:\n",
        "      logits = model(images, training=True)\n",
        "      loss = _loss_fn(logits, labels)\n",
        "    average_loss(loss)\n",
        "    gradients = tape.gradient(loss, model.variables)\n",
        "    optimizer.apply_gradients(zip(gradients, model.variables))\n",
        "  return average_loss.result()\n",
        "   \n",
        "for epoch in range(EPOCHS):\n",
        "  loss = _train_one_epoch(model, training_data)\n",
        "  print(\"Average loss after epoch %d: %.4f\" % (epoch, loss))"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Average loss after epoch 0: 1.0563\n",
            "Average loss after epoch 1: 0.8013\n",
            "Average loss after epoch 2: 0.6306\n",
            "Average loss after epoch 3: 0.5543\n",
            "Average loss after epoch 4: 0.5037\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "uDy1DrYA_2Jz",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "Run the below cell to qualitatively evaluate your model. Note how eager execution interoperates seamlessly with `matplotlib`."
      ]
    },
    {
      "metadata": {
        "id": "vR7rMtpu_3nB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1752
        },
        "outputId": "b212aefa-f4b3-425c-f34d-2491429fa521"
      },
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "\n",
        "sampled_data = test_data.batch(1).shuffle(buffer_size=10000).take(5)\n",
        "for image, label in sampled_data:\n",
        "  plt.figure()\n",
        "  plt.imshow(tf.reshape(image, (28, 28)))\n",
        "  plt.show()\n",
        "  logits = model(image, training=False)\n",
        "  prediction = tf.argmax(logits, axis=1, output_type=tf.int64)\n",
        "  print(\"Prediction: %d\" % prediction)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAKCzFeFbFn4BAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd61cfd1e80>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Prediction: 5\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nlgBg8H/nb4OLnfGqVAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd61bade5c0>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Prediction: 1\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nwIBYAoABsQQAA2IJAAbEEgAM/gepgR0uaefKmwAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd6199ef278>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Prediction: 4\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAEelJREFUeJzt3W9MlfX/x/HXEWJyhg5BIG1ZfR0u\nKr3hhopOE2Q23FxiN0xCdNmGa5pG6hhTtNn8g85NtI0/aS1Z29moG96wILM2dYDKDRu0hrpyzCkC\nkUocDeH8brQfk8R4czyH64DPx624+Hid99nFnl2H61wHl8/n8wkA8J/GOD0AAIwExBIADIglABgQ\nSwAwIJYAYEAsAcCAWAKAAbEEAINwf//h7t27denSJblcLhUUFGjGjBmBnAsAQopfsTx//ryuXbsm\nj8ejq1evqqCgQB6PJ9CzAUDI8OtleE1NjdLT0yVJU6dO1e3bt9XZ2RnQwQAglPgVy7a2Nk2YMKHv\n65iYGLW2tgZsKAAINQG5wMNncQAY7fyKZXx8vNra2vq+vnXrluLi4gI2FACEGr9iOW/ePFVVVUmS\nGhsbFR8fr6ioqIAOBgChxK+r4TNnztSrr76qt99+Wy6XSzt27Aj0XAAQUlx8+C8ADI47eADAgFgC\ngAGxBAADYgkABsQSAAyIJQAYEEsAMCCWAGBALAHAgFgCgAGxBAADYgkABsQSAAyIJQAYEEsAMCCW\nAGBALAHAgFgCgAGxBAADYgkABsQSAAz8+lO4AJz3yy+/PLLtlVdeeWT777//bt7ne++9Z147f/58\n0zqPx2PeZyjjzBIADIglABgQSwAwIJYAYEAsAcCAWAKAAbEEAANiCQAGxBIADIglABi4fD6fz+kh\ngNHsr7/+Mq+tr683r33rrbce2dba2qq4uLh+29rb2837XL16tXntp59+alrndrvN+wxlnFkCgAGx\nBAADYgkABsQSAAyIJQAYEEsAMCCWAGBALAHAgFgCgAF/sAzww/37981rMzMzzWtPnTplXvu4O2O8\nXm+/rysrK837XLJkiXnt2LFjzWtHA84sAcDArzPLuro6bdy4UYmJiZKkadOmafv27QEdDABCid8v\nw2fNmqXi4uJAzgIAIYuX4QBg4Hcsr1y5onXr1mnlypU6d+5cIGcCgJDj1+dZtrS0qL6+XhkZGWpu\nblZOTo6qq6sVERERjBkBwHF+/c4yISGh7y0GU6ZM0cSJE9XS0qLnn38+oMMBoWoobx1aunSpee2T\nvnWos7NTUVFR/bZ9+eWX5n3y1qHH8+tl+IkTJ3T06FFJ/3wyc3t7uxISEgI6GACEEr/OLNPS0rR5\n82b98MMP6u7u1s6dO3kJDmBU8yuWUVFRKikpCfQsABCyuN0ReIj1vcNbtmwx77O7u9u8dii/9//x\nxx8H3P7zzz/3+/p///ufeZ94PN5nCQAGxBIADIglABgQSwAwIJYAYEAsAcCAWAKAAbEEAANiCQAG\nxBIADPz6PEvAaT09Pea1x48fH3D7mjVr9MUXX/Tblpuba9pnb2+v+fE/+eQT89qcnBzz2kmTJpnX\n4slxZgkABsQSAAyIJQAYEEsAMCCWAGBALAHAgFgCgAGxBAADYgkABtzBgxHpcXflDGT16tUDbu/t\n7dWYMf6dL+zcudO8trCw0K/HQGjhzBIADIglABgQSwAwIJYAYEAsAcCAWAKAAbEEAANiCQAGxBIA\nDIglABhwuyNCSnFxsWndRx99ZN7n4/642UC3O77zzjumff77D539l7CwMPNahC7OLAHAgFgCgAGx\nBAADYgkABsQSAAyIJQAYEEsAMCCWAGBALAHAgFgCgAG3OyLovF6vee2kSZNM6+7cuePvOH0Gut2x\npqbG9G9nz579xI+PkcV0ZtnU1KT09HRVVFRIkm7cuKFVq1YpKytLGzdu1N9//x3UIQHAaYPGsqur\nS7t27VJKSkrftuLiYmVlZemrr77SCy+8oMrKyqAOCQBOGzSWERERKi8vV3x8fN+2uro6LVq0SJKU\nmppqfukCACNV+KALwsMVHt5/mdfrVUREhCQpNjZWra2twZkOAELEoLEcDNeHMJjIyEjz2j///DOI\nkzyqt7d3WB8PI5dfsXS73bp3757Gjh2rlpaWfi/RgX/jajhGA7/eZzl37lxVVVVJkqqrqzV//vyA\nDgUAoWbQM8uGhgbt27dP169fV3h4uKqqqnTgwAHl5+fL4/Fo8uTJWrZs2XDMCgCO4U3pCDpehmM0\neOILPHg6ffvtt+a1hw4dMq8NRASfRElJiWkdsXz6cG84ABgQSwAwIJYAYEAsAcCAWAKAAbEEAANi\nCQAGxBIADIglABgQSwAw4HZH+MV6W6D0zydTWU2ZMsW07v79++Z9trS0mNcCj8OZJQAYEEsAMCCW\nAGBALAHAgFgCgAGxBAADYgkABsQSAAyIJQAYEEsAMOB2R/Rz4cKFAbcnJyf3+15tbW1QHv/77783\nrRvKX4FMTk72dxygD2eWAGBALAHAgFgCgAGxBAADYgkABsQSAAyIJQAYEEsAMCCWAGDAHTzoZ8GC\nBQNu93q9/b43lD8YNhTWP1jm9XqD8vjA43BmCQAGxBIADIglABgQSwAwIJYAYEAsAcCAWAKAAbEE\nAANiCQAGxBIADLjd8Slw5MgR89r/uo3R31scZ8yYYV7rcrn8eoxAuXnzpmldV1eXeZ9ut9vfcRBC\nOLMEAANTLJuampSenq6KigpJUn5+vpYuXapVq1Zp1apV+umnn4I5IwA4btCX4V1dXdq1a5dSUlL6\nbc/Ly1NqamrQBgOAUDLomWVERITKy8sVHx8/HPMAQEhy+Xw+n2Xh4cOHNWHCBGVnZys/P1+tra3q\n7u5WbGystm/frpiYmGDPCgCO8etq+Jtvvqno6GglJSWprKxMR44cUWFhYaBnQ4AM5Wr4Bx98MOD2\n3t5ejRnj3/XAoVwNP3/+vGndUK5GP+5/5AM9pzfeeMO0z6+//tr8+FwNHx38+ulPSUlRUlKSJCkt\nLU1NTU0BHQoAQo1fsdywYYOam5slSXV1dUpMTAzoUAAQagZ9Gd7Q0KB9+/bp+vXrCg8PV1VVlbKz\ns7Vp0yZFRkbK7XZrz549wzErADhm0Fi+9tprOn78+CPbrb/bAYDRgNsdnwLt7e2OPv6WLVvMayMi\nIkzrhnKBZyiqqqpM63799VfzPmfOnOnvOAgh3O4IAAbEEgAMiCUAGBBLADAglgBgQCwBwIBYAoAB\nsQQAA2IJAAbEEgAMuN0RfomNjTWvTU5ODvjjnz17NuD7lNT30YODee6554Ly+AhdnFkCgAGxBAAD\nYgkABsQSAAyIJQAYEEsAMCCWAGBALAHAgFgCgAF38MAv48ePN6999tlnA/74FRUVAd+nJM2aNcu0\nLiEhISiPj9DFmSUAGBBLADAglgBgQCwBwIBYAoABsQQAA2IJAAbEEgAMiCUAGBBLADDgdkf45bff\nfjOv/eabb8xrs7OzTet6e3vN+/T5fH59D3gYZ5YAYEAsAcCAWAKAAbEEAANiCQAGxBIADIglABgQ\nSwAwIJYAYEAsAcCA2x0RdO+++25Q1lq5XC6/vgc8zBTLoqIi1dfX68GDB8rNzdX06dO1detW9fT0\nKC4uTvv371dERESwZwUAxwway9raWl2+fFkej0cdHR3KzMxUSkqKsrKylJGRoYMHD6qyslJZWVnD\nMS8AOGLQ31kmJyfr0KFDkqTx48fL6/Wqrq5OixYtkiSlpqaqpqYmuFMCgMMGjWVYWJjcbrckqbKy\nUgsWLJDX6+172R0bG6vW1tbgTgkADjNf4Dl16pQqKyt17NgxLV68uG87nwcY+nbs2BGQtUP5DMmR\nYjQ+JwSHKZZnzpxRSUmJPvvsM40bN05ut1v37t3T2LFj1dLSovj4+GDPiSfw8ccfP/Ha3t5ejRkz\nut5pNtBzWr16tenffv7558EYCSFs0J/+u3fvqqioSKWlpYqOjpYkzZ07V1VVVZKk6upqzZ8/P7hT\nAoDDBj2zPHnypDo6OrRp06a+bXv37tW2bdvk8Xg0efJkLVu2LKhDAoDTBo3lihUrtGLFike28zIE\nwNOEO3ieAnl5eea1Fy5ceOz3lixZ0vffZ8+eNe/zzp075rVAqBpdv7EHgCAhlgBgQCwBwIBYAoAB\nsQQAA2IJAAbEEgAMiCUAGBBLADAglgBg4PLxgZTww3fffWde+/Btkk543I+4z+d75A+W1dbWmvY5\ne/bsJ54LIwtnlgBgQCwBwIBYAoABsQQAA2IJAAbEEgAMiCUAGBBLADAglgBgQCwBwIDbHQHAgDNL\nADAglgBgQCwBwIBYAoABsQQAA2IJAAbEEgAMiCUAGBBLADAglgBgQCwBwIBYAoABsQQAA2IJAAbE\nEgAMiCUAGBBLADAglgBgQCwBwIBYAoABsQQAg3DLoqKiItXX1+vBgwfKzc3V6dOn1djYqOjoaEnS\n2rVrtXDhwmDOCQCOGjSWtbW1unz5sjwejzo6OpSZmak5c+YoLy9PqampwzEjADhu0FgmJydrxowZ\nkqTx48fL6/Wqp6cn6IMBQChx+Xw+n3Wxx+PRxYsXFRYWptbWVnV3dys2Nlbbt29XTExMMOcEAEeZ\nY3nq1CmVlpbq2LFjamhoUHR0tJKSklRWVqabN2+qsLAw2LMCgGNMV8PPnDmjkpISlZeXa9y4cUpJ\nSVFSUpIkKS0tTU1NTUEdEgCcNmgs7969q6KiIpWWlvZd/d6wYYOam5slSXV1dUpMTAzulADgsEEv\n8Jw8eVIdHR3atGlT37bly5dr06ZNioyMlNvt1p49e4I6JAA4bUgXeADgacUdPABgQCwBwIBYAoAB\nsQQAA2IJAAbEEgAMiCUAGBBLADAglgBgQCwBwIBYAoABsQQAA2IJAAbEEgAMiCUAGBBLADAglgBg\nQCwBwIBYAoABsQQAA2IJAAbEEgAMiCUAGBBLADAglgBgQCwBwIBYAoABsQQAA2IJAAbhTjzo7t27\ndenSJblcLhUUFGjGjBlOjBFQdXV12rhxoxITEyVJ06ZN0/bt2x2eyn9NTU16//33tWbNGmVnZ+vG\njRvaunWrenp6FBcXp/379ysiIsLpMYfk388pPz9fjY2Nio6OliStXbtWCxcudHbIISoqKlJ9fb0e\nPHig3NxcTZ8+fcQfJ+nR53X69GnHj9Wwx/L8+fO6du2aPB6Prl69qoKCAnk8nuEeIyhmzZql4uJi\np8d4Yl1dXdq1a5dSUlL6thUXFysrK0sZGRk6ePCgKisrlZWV5eCUQzPQc5KkvLw8paamOjTVk6mt\nrdXly5fl8XjU0dGhzMxMpaSkjOjjJA38vObMmeP4sRr2l+E1NTVKT0+XJE2dOlW3b99WZ2fncI+B\n/xAREaHy8nLFx8f3baurq9OiRYskSampqaqpqXFqPL8M9JxGuuTkZB06dEiSNH78eHm93hF/nKSB\nn1dPT4/DUzkQy7a2Nk2YMKHv65iYGLW2tg73GEFx5coVrVu3TitXrtS5c+ecHsdv4eHhGjt2bL9t\nXq+37+VcbGzsiDtmAz0nSaqoqFBOTo4+/PBD/fHHHw5M5r+wsDC53W5JUmVlpRYsWDDij5M08PMK\nCwtz/Fg58jvLh/l8PqdHCIgXX3xR69evV0ZGhpqbm5WTk6Pq6uoR+fuiwYyWY/bmm28qOjpaSUlJ\nKisr05EjR1RYWOj0WEN26tQpVVZW6tixY1q8eHHf9pF+nB5+Xg0NDY4fq2E/s4yPj1dbW1vf17du\n3VJcXNxwjxFwCQkJWrJkiVwul6ZMmaKJEyeqpaXF6bECxu126969e5KklpaWUfFyNiUlRUlJSZKk\ntLQ0NTU1OTzR0J05c0YlJSUqLy/XuHHjRs1x+vfzCoVjNeyxnDdvnqqqqiRJjY2Nio+PV1RU1HCP\nEXAnTpzQ0aNHJUmtra1qb29XQkKCw1MFzty5c/uOW3V1tebPn+/wRE9uw4YNam5ulvTP72T//50M\nI8Xdu3dVVFSk0tLSvqvEo+E4DfS8QuFYuXwOnKsfOHBAFy9elMvl0o4dO/Tyyy8P9wgB19nZqc2b\nN+vOnTvq7u7W+vXr9frrrzs9ll8aGhq0b98+Xb9+XeHh4UpISNCBAweUn5+v+/fva/LkydqzZ4+e\neeYZp0c1G+g5ZWdnq6ysTJGRkXK73dqzZ49iY2OdHtXM4/Ho8OHDeumll/q27d27V9u2bRuxx0ka\n+HktX75cFRUVjh4rR2IJACMNd/AAgAGxBAADYgkABsQSAAyIJQAYEEsAMCCWAGBALAHA4P8ALqDX\nN3rmU3AAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd62944c6d8>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Prediction: 1\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUsAAAFKCAYAAACU6307AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\niCUAGBBLADAglgBg8B9OkjtgR8VvdgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<matplotlib.figure.Figure at 0x7fd619a40b00>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Prediction: 6\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "4SJizeJtNaAs",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "# Profiling\n",
        "\n",
        "If you want to drill down into the performance characteristics of your code, you can use native Python profilers like [`cProfile`](https://docs.python.org/3/library/profile.html). In the next exercise, you'll do just that."
      ]
    },
    {
      "metadata": {
        "id": "_2v0QnG8__PJ",
        "colab_type": "text"
      },
      "cell_type": "markdown",
      "source": [
        "## Exercise!\n",
        "\n",
        "This exercise does not require coding. If you have not completed the training exercise, replace `train_one_epoch` below with `_train_one_epoch`.\n",
        "\n",
        "Run the below cell and inspect the printed profiles. What parts of the code appear to be hotspots or\n",
        "bottlenecks? How does sorting the profile by total time compare to sorting it\n",
        "by cumulative time?\n",
        "\n"
      ]
    },
    {
      "metadata": {
        "id": "IFypaYbG_9fB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 714
        },
        "outputId": "d9c3596b-a165-4edd-fc6b-53ccd0d01d19"
      },
      "cell_type": "code",
      "source": [
        "import cProfile\n",
        "import pstats\n",
        "\n",
        "cProfile.run(\"train_one_epoch(model, training_data, optimizer)\", \"training_profile\")\n",
        "\n",
        "stats = pstats.Stats(\"training_profile\").strip_dirs().sort_stats(\"tottime\")\n",
        "stats.print_stats(10)\n",
        "\n",
        "stats.sort_stats(\"cumtime\").print_stats(10)"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Thu Jun  7 12:25:04 2018    training_profile\n",
            "\n",
            "         92209 function calls (91817 primitive calls) in 3.446 seconds\n",
            "\n",
            "   Ordered by: internal time\n",
            "   List reduced from 672 to 10 due to restriction <10>\n",
            "\n",
            "   ncalls  tottime  percall  cumtime  percall filename:lineno(function)\n",
            "     1080    2.552    0.002    2.552    0.002 {built-in method _pywrap_tensorflow_internal.TFE_Py_FastPathExecute}\n",
            "       83    0.753    0.009    0.753    0.009 {built-in method _pywrap_tensorflow_internal.TFE_Py_Execute}\n",
            "       16    0.006    0.000    1.019    0.064 network.py:736(_run_internal_graph)\n",
            "       16    0.005    0.000    2.253    0.141 {built-in method _pywrap_tensorflow_internal.TFE_Py_TapeGradient}\n",
            "     2321    0.004    0.000    0.007    0.000 abc.py:178(__instancecheck__)\n",
            "      288    0.004    0.000    0.009    0.000 inspect.py:2092(_signature_from_function)\n",
            "      878    0.004    0.000    0.005    0.000 ops.py:5936(__enter__)\n",
            "      288    0.004    0.000    0.016    0.000 inspect.py:1079(getfullargspec)\n",
            "    11006    0.003    0.000    0.005    0.000 {built-in method builtins.isinstance}\n",
            "      768    0.003    0.000    0.008    0.000 {built-in method _pywrap_tensorflow_internal.Flatten}\n",
            "\n",
            "\n",
            "Thu Jun  7 12:25:04 2018    training_profile\n",
            "\n",
            "         92209 function calls (91817 primitive calls) in 3.446 seconds\n",
            "\n",
            "   Ordered by: cumulative time\n",
            "   List reduced from 672 to 10 due to restriction <10>\n",
            "\n",
            "   ncalls  tottime  percall  cumtime  percall filename:lineno(function)\n",
            "        1    0.000    0.000    3.446    3.446 {built-in method builtins.exec}\n",
            "        1    0.000    0.000    3.446    3.446 <string>:1(<module>)\n",
            "        1    0.001    0.001    3.446    3.446 <ipython-input-14-bcffed60b545>:9(train_one_epoch)\n",
            "     1080    2.552    0.002    2.552    0.002 {built-in method _pywrap_tensorflow_internal.TFE_Py_FastPathExecute}\n",
            "       16    0.000    0.000    2.255    0.141 backprop.py:739(gradient)\n",
            "       16    0.000    0.000    2.253    0.141 imperative_grad.py:31(imperative_grad)\n",
            "       16    0.005    0.000    2.253    0.141 {built-in method _pywrap_tensorflow_internal.TFE_Py_TapeGradient}\n",
            "      400    0.002    0.000    2.246    0.006 backprop.py:145(grad_fn)\n",
            "      400    0.002    0.000    2.239    0.006 backprop.py:95(_magic_gradient_function)\n",
            "       32    0.001    0.000    1.601    0.050 nn_grad.py:497(_Conv2DGrad)\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<pstats.Stats at 0x7fd61f841710>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 17
        }
      ]
    },
    {
      "metadata": {
        "id": "8ixpnyCNNTI4",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}