notes-2018-03-02.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# UBCS3 Machine Learning:<br /> Hands on Machine Learning with Scikit-Learn and Tensorflow\n",
    "\n",
    "## Chapter 11: Training Deep Neural Networks (2 March 2018)\n",
    "\n",
    "See [Hands on Machine Learning with Scikit-Learn and Tensorflow](https://github.com/ageron/handson-ml) by Aurélien Géron."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Goals\n",
    "\n",
    "* Avoid vanishing gradients / exploding gradients problem.\n",
    "* Expedite training for large networks.\n",
    "* Avoid overfitting large parameter sets to training data.\n",
    "\n",
    "### Important Notes\n",
    "\n",
    "* Almost all of this material is compiled directly from Géron's [book](http://shop.oreilly.com/product/0636920052289.do) and/or [IPython Notebook](https://github.com/ageron/handson-ml/blob/master/11_deep_learning.ipynb), with the aim of better accommodating time constraints [at the expense of greater leaps between techniques]."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import os\n",
    "\n",
    "# to make this notebook's output stable across runs\n",
    "def reset_graph(seed=42):\n",
    "    tf.reset_default_graph()\n",
    "    tf.set_random_seed(seed)\n",
    "    np.random.seed(seed)\n",
    "\n",
    "# To plot pretty figures\n",
    "%matplotlib inline\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "plt.rcParams['axes.labelsize'] = 14\n",
    "plt.rcParams['xtick.labelsize'] = 12\n",
    "plt.rcParams['ytick.labelsize'] = 12"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/aberk/anaconda/lib/python3.5/site-packages/h5py/__init__.py:34: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "reset_graph()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Vanishing Gradients\n",
    "\n",
    "<img src=\"./img/sigmoid.png\" width=\"400px\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Xavier/Glorot and He Weight initialization\n",
    "\n",
    "For the signal to flow properly, the authors argue that we need the variance of the outputs of each layer to be equal to the variance of its inputs, and we also need the gradients to have equal variance before and after flowing through a layer in the reverse direction. A good compromise that has proven to work in practice: the connection weights must be initialized randomly as\n",
    "$$\n",
    "w \\sim \\mathcal N \\big( 0, \\frac{2}{n_{\\text{in}} + n_{\\text{out}}}\\big) %\n",
    "\\qquad \\text{or} \\qquad %\n",
    "w \\sim \\mathrm{Unif}\\big(\\big[ - \\sqrt{\\frac{6}{n_{\\text{in}} + n_{\\text{out}}}},  \\sqrt{\\frac{6}{n_{\\text{in}} + n_{\\text{out}}}}\\big] \\big)\n",
    "$$\n",
    "\n",
    "| Activation Function | Uniform distribution $[-r,r]$ | Normal distribution                |\n",
    "| ------------------- | ----------------------------- | ---------------------------------- |\n",
    "| Logistic            | $r = \\sqrt{\\frac{6}{n_{\\text{in}} + n_{\\text{out}}}}$ | $\\sigma = \\sqrt{\\frac{2}{n_{\\text{in}} + n_{\\text{out}}}}$ |\n",
    "| Hyperbolic Tangent  | $r = 4 \\sqrt{\\frac{6}{n_{\\text{in}} + n_{\\text{out}}}}$ | $\\sigma = 4\\sqrt{\\frac{2}{n_{\\text{in}} + n_{\\text{out}}}}$ | \n",
    "| ReLU (and variants) | $r = \\sqrt 2 \\sqrt{\\frac{6}{n_{\\text{in}} + n_{\\text{out}}}}$ | $\\sigma = \\sqrt{2}\\sqrt{\\frac{2}{n_{\\text{in}} + n_{\\text{out}}}}$ |\n",
    "\n",
    "Ref: [(Glorot & Bengio, 2010)](http://proceedings.mlr.press/v9/glorot10a/glorot10a.pdf)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_inputs = 28 * 28  # MNIST\n",
    "n_hidden1 = 300\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "he_init = tf.contrib.layers.variance_scaling_initializer() # for ReLU\n",
    "xav_init = tf.contrib.layers.xavier_initializer() # for logistic # not used\n",
    "hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.relu,\n",
    "                          kernel_initializer=he_init, name=\"hidden1\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Dying ReLUs\n",
    "\n",
    "During training, some ReLU neurons \"die\", outputting only $0$. Sometimes, more than half of a network's neurons can die during training (esp. if large learning rate); unlikely for them to \"come back to life\" because the gradient is $0$! Some possible solutions are below.\n",
    "\n",
    "#### Solution: Pick activation function that won't die\n",
    "\n",
    "* **Leaky ReLU:** Defined by $\\text{LReLU}(z) := \\max\\{ \\alpha z , z \\}, \\alpha \\approx .15$\n",
    "\n",
    "\n",
    "<img src=\"./img/lrelu.png\" width=\"400px\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Implementing Leaky ReLU in `tensorflow`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Leaky ReLU is not built in to `tensorflow` so we have to make it ourselves."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "\n",
    "def leaky_relu(z, name=None):\n",
    "    return tf.maximum(0.01 * z, z, name=name)\n",
    "\n",
    "hidden1 = tf.layers.dense(X, n_hidden1, activation=leaky_relu, name=\"hidden1\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's train a neural network on MNIST using the Leaky ReLU. First let's create the graph:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "n_inputs = 28 * 28  # MNIST\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 100\n",
    "n_outputs = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, activation=leaky_relu, name=\"hidden1\")\n",
    "    hidden2 = tf.layers.dense(hidden1, n_hidden2, activation=leaky_relu, name=\"hidden2\")\n",
    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits)\n",
    "    loss = tf.reduce_mean(xentropy, name=\"loss\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learning_rate = 0.01\n",
    "\n",
    "with tf.name_scope(\"train\"):\n",
    "    optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "    training_op = optimizer.minimize(loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"eval\"):\n",
    "    correct = tf.nn.in_top_k(logits, y, 1)\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32), name='accuracy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's load the data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting /tmp/data/train-images-idx3-ubyte.gz\n",
      "Extracting /tmp/data/train-labels-idx1-ubyte.gz\n",
      "Extracting /tmp/data/t10k-images-idx3-ubyte.gz\n",
      "Extracting /tmp/data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets(\"/tmp/data/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_epochs = 40\n",
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Batch accuracy: 0.86 Validation accuracy: 0.9044\n",
      "5 Batch accuracy: 0.94 Validation accuracy: 0.951\n",
      "10 Batch accuracy: 0.96 Validation accuracy: 0.9666\n",
      "15 Batch accuracy: 1.0 Validation accuracy: 0.9722\n",
      "20 Batch accuracy: 1.0 Validation accuracy: 0.9748\n",
      "25 Batch accuracy: 1.0 Validation accuracy: 0.9768\n",
      "30 Batch accuracy: 0.98 Validation accuracy: 0.9778\n",
      "35 Batch accuracy: 0.96 Validation accuracy: 0.9796\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        if epoch % 5 == 0:\n",
    "            acc_train = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
    "            acc_test = accuracy.eval(feed_dict={X: mnist.validation.images, y: mnist.validation.labels})\n",
    "            print(epoch, \"Batch accuracy:\", acc_train, \"Validation accuracy:\", acc_test)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* **Exponential Linear Unit:** Defined by $\\text{ELU}_\\alpha(z) := \\begin{cases}\\alpha (e^z - 1) & z < 0\\\\ z & z \\geq 0\\end{cases}$\n",
    "\n",
    "<img src=\"./img/elu.png\" width=\"400px\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "ELUs are easy to implement in `tensorflow` because they're part of the default package."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.elu, name=\"hidden1\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Batch Normalization\n",
    "\n",
    "The technique consists of adding an operation in the model just before the activation function of each layer, simply zero-centering and normalizing the inputs, then scaling and shifting the result using two new parameters per layer (one for scaling, the other for shifting). BN lets the model learn the optimal scale and mean of the inputs for each layer.\n",
    "\n",
    "It estimates the inputs’ mean and standard deviation by evaluating the mean and standard deviation of the inputs over the current mini-batch:\n",
    "\\begin{align*}\n",
    "\\text{empirical mean over $B$} = \\mu_B &:= \\frac{1}{m_B} \\sum_{i=1}^{m_B} x^{(i)}\\\\\n",
    "\\text{empirical standard deviation over $B$} = \\sigma^2_B &:= \\frac{1}{m_B} \\sum_{i=1}^{m_B} (x^{(i)} - \\mu_B)^2\\\\\n",
    "\\text{zero-centered and normalized input} = \\hat x^{(i)} &:= \\frac{x^{(i)} - \\mu_B}{\\sqrt{\\sigma^2_B + \\varepsilon}}\\\\\n",
    "\\text{output of the BN operation} = z^{(i)} &:= \\gamma \\hat x^{(i)} + \\beta\n",
    "\\end{align*}\n",
    "Above, $m_B$ is the number of instances in the mini-batch, $\\gamma$ is a scaling parameter for the layer, $\\beta$ is a shifting/offset parameter for the layer, $\\varepsilon$ is a tiny positive number to avoid division by zero ($10^{-5}$).\n",
    "\n",
    "The BN algorithm uses exponential decay to compute the running averages over consecutive mini-batches:\n",
    "$$\n",
    "\\hat \\nu \\leftarrow \\hat \\nu \\cdot \\text{momentum} + \\nu \\cdot (1 - \\text{momentum})\n",
    "$$\n",
    "where $\\text{momentum} \\approx 0.9$.\n",
    "\n",
    "**Note:** training is rather slow at first while Gradient Descent is searching for the optimal scales and offsets for each layer, but it accelerates once it has found reasonably good values\n",
    "\n",
    "Ref: [(Ioffe & Szegedy, 2015)](https://arxiv.org/pdf/1502.03167v3.pdf)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Implementing Batch Normalization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First reset the workspace and define some parameters for the network we'll build."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "import tensorflow as tf\n",
    "\n",
    "n_inputs = 28 * 28\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 100\n",
    "n_outputs = 10\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define a `placeholder_with_default` to tell `tensorflow` whether or not we're in `'training'` mode. We only want to adjust batch normalization while in `'training'` mode. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "training = tf.placeholder_with_default(False, shape=(), name='training')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then construct a network! `tf.layers` has a `batch_normalization` function that takes a `momentum` parameter, as described above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from functools import partial\n",
    "\n",
    "my_batch_norm_layer = partial(tf.layers.batch_normalization,\n",
    "                              training=training, momentum=0.9)\n",
    "\n",
    "hidden1 = tf.layers.dense(X, n_hidden1, name=\"hidden1\")\n",
    "bn1 = my_batch_norm_layer(hidden1)\n",
    "bn1_act = tf.nn.elu(bn1)\n",
    "hidden2 = tf.layers.dense(bn1_act, n_hidden2, name=\"hidden2\")\n",
    "bn2 = my_batch_norm_layer(hidden2)\n",
    "bn2_act = tf.nn.elu(bn2)\n",
    "logits_before_bn = tf.layers.dense(bn2_act, n_outputs, name=\"outputs\")\n",
    "logits = my_batch_norm_layer(logits_before_bn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Example:** Build a network for MNIST using ELU activation and BN at each layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "batch_norm_momentum = 0.9\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")\n",
    "training = tf.placeholder_with_default(False, shape=(), name='training')\n",
    "\n",
    "with tf.name_scope(\"dnn\"):\n",
    "    he_init = tf.contrib.layers.variance_scaling_initializer()\n",
    "\n",
    "    my_batch_norm_layer = partial(\n",
    "            tf.layers.batch_normalization,\n",
    "            training=training,\n",
    "            momentum=batch_norm_momentum)\n",
    "\n",
    "    my_dense_layer = partial(\n",
    "            tf.layers.dense,\n",
    "            kernel_initializer=he_init)\n",
    "\n",
    "    hidden1 = my_dense_layer(X, n_hidden1, name=\"hidden1\")\n",
    "    bn1 = tf.nn.elu(my_batch_norm_layer(hidden1))\n",
    "    hidden2 = my_dense_layer(bn1, n_hidden2, name=\"hidden2\")\n",
    "    bn2 = tf.nn.elu(my_batch_norm_layer(hidden2))\n",
    "    logits_before_bn = my_dense_layer(bn2, n_outputs, name=\"outputs\")\n",
    "    logits = my_batch_norm_layer(logits_before_bn)\n",
    "\n",
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits)\n",
    "    loss = tf.reduce_mean(xentropy, name=\"loss\")\n",
    "\n",
    "with tf.name_scope(\"train\"):\n",
    "    optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "    training_op = optimizer.minimize(loss)\n",
    "\n",
    "with tf.name_scope(\"eval\"):\n",
    "    correct = tf.nn.in_top_k(logits, y, 1)\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32), \n",
    "                              name='accuracy')\n",
    "    \n",
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_epochs = 20\n",
    "batch_size = 200"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<tf.Tensor 'dnn/batch_normalization/AssignMovingAvg:0' shape=(300,) dtype=float32_ref>,\n",
       " <tf.Tensor 'dnn/batch_normalization/AssignMovingAvg_1:0' shape=(300,) dtype=float32_ref>,\n",
       " <tf.Tensor 'dnn/batch_normalization_2/AssignMovingAvg:0' shape=(100,) dtype=float32_ref>,\n",
       " <tf.Tensor 'dnn/batch_normalization_2/AssignMovingAvg_1:0' shape=(100,) dtype=float32_ref>,\n",
       " <tf.Tensor 'dnn/batch_normalization_3/AssignMovingAvg:0' shape=(10,) dtype=float32_ref>,\n",
       " <tf.Tensor 'dnn/batch_normalization_3/AssignMovingAvg_1:0' shape=(10,) dtype=float32_ref>]"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.get_collection(tf.GraphKeys.UPDATE_OPS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Test accuracy: 0.8665\n",
      "1 Test accuracy: 0.8963\n",
      "2 Test accuracy: 0.9128\n",
      "3 Test accuracy: 0.9229\n",
      "4 Test accuracy: 0.9277\n",
      "5 Test accuracy: 0.9344\n",
      "6 Test accuracy: 0.9382\n",
      "7 Test accuracy: 0.9422\n",
      "8 Test accuracy: 0.9445\n",
      "9 Test accuracy: 0.947\n",
      "10 Test accuracy: 0.9504\n",
      "11 Test accuracy: 0.9536\n",
      "12 Test accuracy: 0.9571\n",
      "13 Test accuracy: 0.9568\n",
      "14 Test accuracy: 0.96\n",
      "15 Test accuracy: 0.9599\n",
      "16 Test accuracy: 0.9622\n",
      "17 Test accuracy: 0.9619\n",
      "18 Test accuracy: 0.9636\n",
      "19 Test accuracy: 0.9638\n"
     ]
    }
   ],
   "source": [
    "extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
    "            sess.run([training_op, extra_update_ops],\n",
    "                     feed_dict={training: True, X: X_batch, y: y_batch})\n",
    "        accuracy_val = accuracy.eval(feed_dict={X: mnist.test.images,\n",
    "                                                y: mnist.test.labels})\n",
    "        print(epoch, \"Test accuracy:\", accuracy_val)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is worse accuracy than before! The Géron has only claimed that ELU and BN perform well for deep nets, of which our toy is not an example. It is suggested this is why we cannot expect great performance out of them in this example. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**One more thing:** notice that the list of trainable variables is shorter than the list of all global variables. This is because the moving averages are non-trainable variables. If you want to reuse a pretrained neural network (see below), you must not forget these non-trainable variables."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['hidden1/kernel:0',\n",
       " 'hidden1/bias:0',\n",
       " 'batch_normalization/gamma:0',\n",
       " 'batch_normalization/beta:0',\n",
       " 'hidden2/kernel:0',\n",
       " 'hidden2/bias:0',\n",
       " 'batch_normalization_1/gamma:0',\n",
       " 'batch_normalization_1/beta:0',\n",
       " 'outputs/kernel:0',\n",
       " 'outputs/bias:0',\n",
       " 'batch_normalization_2/gamma:0',\n",
       " 'batch_normalization_2/beta:0']"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[v.name for v in tf.trainable_variables()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['hidden1/kernel:0',\n",
       " 'hidden1/bias:0',\n",
       " 'batch_normalization/gamma:0',\n",
       " 'batch_normalization/beta:0',\n",
       " 'batch_normalization/moving_mean:0',\n",
       " 'batch_normalization/moving_variance:0',\n",
       " 'hidden2/kernel:0',\n",
       " 'hidden2/bias:0',\n",
       " 'batch_normalization_1/gamma:0',\n",
       " 'batch_normalization_1/beta:0',\n",
       " 'batch_normalization_1/moving_mean:0',\n",
       " 'batch_normalization_1/moving_variance:0',\n",
       " 'outputs/kernel:0',\n",
       " 'outputs/bias:0',\n",
       " 'batch_normalization_2/gamma:0',\n",
       " 'batch_normalization_2/beta:0',\n",
       " 'batch_normalization_2/moving_mean:0',\n",
       " 'batch_normalization_2/moving_variance:0']"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[v.name for v in tf.global_variables()]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reusing pretrained layers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"./img/pretrained-layers.png\" width=\"400px\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reusing a Tensorflow Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that by default, a `Saver` saves the structure of the graph into a `.meta` file, so that's the file you should load:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "saver = tf.train.import_meta_graph(\"./my_model_final.ckpt.meta\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next you need to get a handle on all the operations you will need for training. If you don't know the graph's structure, you can list all the operations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X\n",
      "y\n",
      "hidden1/kernel/Initializer/random_uniform/shape\n",
      "hidden1/kernel/Initializer/random_uniform/min\n",
      "hidden1/kernel/Initializer/random_uniform/max\n",
      "hidden1/kernel/Initializer/random_uniform/RandomUniform\n",
      "hidden1/kernel/Initializer/random_uniform/sub\n",
      "hidden1/kernel/Initializer/random_uniform/mul\n",
      "hidden1/kernel/Initializer/random_uniform\n",
      "hidden1/kernel\n",
      "hidden1/kernel/Assign\n",
      "hidden1/kernel/read\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/mul\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum/reduction_indices\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/Rsqrt\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1/y\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/Const\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv/y\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/Minimum\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_2\n",
      "dnn/hidden1/kernel/Regularizer/clip_by_norm\n",
      "dnn/hidden1/kernel/Regularizer/max_norm\n",
      "hidden1/bias/Initializer/zeros\n",
      "hidden1/bias\n",
      "hidden1/bias/Assign\n",
      "hidden1/bias/read\n",
      "dnn/hidden1/MatMul\n",
      "dnn/hidden1/BiasAdd\n",
      "dnn/hidden1/Relu\n",
      "hidden2/kernel/Initializer/random_uniform/shape\n",
      "hidden2/kernel/Initializer/random_uniform/min\n",
      "hidden2/kernel/Initializer/random_uniform/max\n",
      "hidden2/kernel/Initializer/random_uniform/RandomUniform\n",
      "hidden2/kernel/Initializer/random_uniform/sub\n",
      "hidden2/kernel/Initializer/random_uniform/mul\n",
      "hidden2/kernel/Initializer/random_uniform\n",
      "hidden2/kernel\n",
      "hidden2/kernel/Assign\n",
      "hidden2/kernel/read\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/mul\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum/reduction_indices\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/Rsqrt\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1/y\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/Const\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv/y\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/Minimum\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_2\n",
      "dnn/hidden2/kernel/Regularizer/clip_by_norm\n",
      "dnn/hidden2/kernel/Regularizer/max_norm\n",
      "hidden2/bias/Initializer/zeros\n",
      "hidden2/bias\n",
      "hidden2/bias/Assign\n",
      "hidden2/bias/read\n",
      "dnn/hidden2/MatMul\n",
      "dnn/hidden2/BiasAdd\n",
      "dnn/hidden2/Relu\n",
      "outputs/kernel/Initializer/random_uniform/shape\n",
      "outputs/kernel/Initializer/random_uniform/min\n",
      "outputs/kernel/Initializer/random_uniform/max\n",
      "outputs/kernel/Initializer/random_uniform/RandomUniform\n",
      "outputs/kernel/Initializer/random_uniform/sub\n",
      "outputs/kernel/Initializer/random_uniform/mul\n",
      "outputs/kernel/Initializer/random_uniform\n",
      "outputs/kernel\n",
      "outputs/kernel/Assign\n",
      "outputs/kernel/read\n",
      "outputs/bias/Initializer/zeros\n",
      "outputs/bias\n",
      "outputs/bias/Assign\n",
      "outputs/bias/read\n",
      "dnn/outputs/MatMul\n",
      "dnn/outputs/BiasAdd\n",
      "loss/SparseSoftmaxCrossEntropyWithLogits/Shape\n",
      "loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits\n",
      "loss/Const\n",
      "loss/loss\n",
      "train/gradients/Shape\n",
      "train/gradients/Const\n",
      "train/gradients/Fill\n",
      "train/gradients/loss/loss_grad/Reshape/shape\n",
      "train/gradients/loss/loss_grad/Reshape\n",
      "train/gradients/loss/loss_grad/Shape\n",
      "train/gradients/loss/loss_grad/Tile\n",
      "train/gradients/loss/loss_grad/Shape_1\n",
      "train/gradients/loss/loss_grad/Shape_2\n",
      "train/gradients/loss/loss_grad/Const\n",
      "train/gradients/loss/loss_grad/Prod\n",
      "train/gradients/loss/loss_grad/Const_1\n",
      "train/gradients/loss/loss_grad/Prod_1\n",
      "train/gradients/loss/loss_grad/Maximum/y\n",
      "train/gradients/loss/loss_grad/Maximum\n",
      "train/gradients/loss/loss_grad/floordiv\n",
      "train/gradients/loss/loss_grad/Cast\n",
      "train/gradients/loss/loss_grad/truediv\n",
      "train/gradients/zeros_like\n",
      "train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/PreventGradient\n",
      "train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims/dim\n",
      "train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims\n",
      "train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul\n",
      "train/gradients/dnn/outputs/BiasAdd_grad/BiasAddGrad\n",
      "train/gradients/dnn/outputs/BiasAdd_grad/tuple/group_deps\n",
      "train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency\n",
      "train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency_1\n",
      "train/gradients/dnn/outputs/MatMul_grad/MatMul\n",
      "train/gradients/dnn/outputs/MatMul_grad/MatMul_1\n",
      "train/gradients/dnn/outputs/MatMul_grad/tuple/group_deps\n",
      "train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency\n",
      "train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency_1\n",
      "train/gradients/dnn/hidden2/Relu_grad/ReluGrad\n",
      "train/gradients/dnn/hidden2/BiasAdd_grad/BiasAddGrad\n",
      "train/gradients/dnn/hidden2/BiasAdd_grad/tuple/group_deps\n",
      "train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency\n",
      "train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency_1\n",
      "train/gradients/dnn/hidden2/MatMul_grad/MatMul\n",
      "train/gradients/dnn/hidden2/MatMul_grad/MatMul_1\n",
      "train/gradients/dnn/hidden2/MatMul_grad/tuple/group_deps\n",
      "train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency\n",
      "train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency_1\n",
      "train/gradients/dnn/hidden1/Relu_grad/ReluGrad\n",
      "train/gradients/dnn/hidden1/BiasAdd_grad/BiasAddGrad\n",
      "train/gradients/dnn/hidden1/BiasAdd_grad/tuple/group_deps\n",
      "train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency\n",
      "train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency_1\n",
      "train/gradients/dnn/hidden1/MatMul_grad/MatMul\n",
      "train/gradients/dnn/hidden1/MatMul_grad/MatMul_1\n",
      "train/gradients/dnn/hidden1/MatMul_grad/tuple/group_deps\n",
      "train/gradients/dnn/hidden1/MatMul_grad/tuple/control_dependency\n",
      "train/gradients/dnn/hidden1/MatMul_grad/tuple/control_dependency_1\n",
      "hidden1/kernel/Momentum/Initializer/zeros\n",
      "hidden1/kernel/Momentum\n",
      "hidden1/kernel/Momentum/Assign\n",
      "hidden1/kernel/Momentum/read\n",
      "hidden1/bias/Momentum/Initializer/zeros\n",
      "hidden1/bias/Momentum\n",
      "hidden1/bias/Momentum/Assign\n",
      "hidden1/bias/Momentum/read\n",
      "hidden2/kernel/Momentum/Initializer/zeros\n",
      "hidden2/kernel/Momentum\n",
      "hidden2/kernel/Momentum/Assign\n",
      "hidden2/kernel/Momentum/read\n",
      "hidden2/bias/Momentum/Initializer/zeros\n",
      "hidden2/bias/Momentum\n",
      "hidden2/bias/Momentum/Assign\n",
      "hidden2/bias/Momentum/read\n",
      "outputs/kernel/Momentum/Initializer/zeros\n",
      "outputs/kernel/Momentum\n",
      "outputs/kernel/Momentum/Assign\n",
      "outputs/kernel/Momentum/read\n",
      "outputs/bias/Momentum/Initializer/zeros\n",
      "outputs/bias/Momentum\n",
      "outputs/bias/Momentum/Assign\n",
      "outputs/bias/Momentum/read\n",
      "train/Momentum/learning_rate\n",
      "train/Momentum/momentum\n",
      "train/Momentum/update_hidden1/kernel/ApplyMomentum\n",
      "train/Momentum/update_hidden1/bias/ApplyMomentum\n",
      "train/Momentum/update_hidden2/kernel/ApplyMomentum\n",
      "train/Momentum/update_hidden2/bias/ApplyMomentum\n",
      "train/Momentum/update_outputs/kernel/ApplyMomentum\n",
      "train/Momentum/update_outputs/bias/ApplyMomentum\n",
      "train/Momentum\n",
      "eval/in_top_k/InTopKV2/k\n",
      "eval/in_top_k/InTopKV2\n",
      "eval/Cast\n",
      "eval/Const\n",
      "eval/Mean\n",
      "init\n",
      "save/Const\n",
      "save/SaveV2/tensor_names\n",
      "save/SaveV2/shape_and_slices\n",
      "save/SaveV2\n",
      "save/control_dependency\n",
      "save/RestoreV2/tensor_names\n",
      "save/RestoreV2/shape_and_slices\n",
      "save/RestoreV2\n",
      "save/Assign\n",
      "save/RestoreV2_1/tensor_names\n",
      "save/RestoreV2_1/shape_and_slices\n",
      "save/RestoreV2_1\n",
      "save/Assign_1\n",
      "save/RestoreV2_2/tensor_names\n",
      "save/RestoreV2_2/shape_and_slices\n",
      "save/RestoreV2_2\n",
      "save/Assign_2\n",
      "save/RestoreV2_3/tensor_names\n",
      "save/RestoreV2_3/shape_and_slices\n",
      "save/RestoreV2_3\n",
      "save/Assign_3\n",
      "save/RestoreV2_4/tensor_names\n",
      "save/RestoreV2_4/shape_and_slices\n",
      "save/RestoreV2_4\n",
      "save/Assign_4\n",
      "save/RestoreV2_5/tensor_names\n",
      "save/RestoreV2_5/shape_and_slices\n",
      "save/RestoreV2_5\n",
      "save/Assign_5\n",
      "save/RestoreV2_6/tensor_names\n",
      "save/RestoreV2_6/shape_and_slices\n",
      "save/RestoreV2_6\n",
      "save/Assign_6\n",
      "save/RestoreV2_7/tensor_names\n",
      "save/RestoreV2_7/shape_and_slices\n",
      "save/RestoreV2_7\n",
      "save/Assign_7\n",
      "save/RestoreV2_8/tensor_names\n",
      "save/RestoreV2_8/shape_and_slices\n",
      "save/RestoreV2_8\n",
      "save/Assign_8\n",
      "save/RestoreV2_9/tensor_names\n",
      "save/RestoreV2_9/shape_and_slices\n",
      "save/RestoreV2_9\n",
      "save/Assign_9\n",
      "save/RestoreV2_10/tensor_names\n",
      "save/RestoreV2_10/shape_and_slices\n",
      "save/RestoreV2_10\n",
      "save/Assign_10\n",
      "save/RestoreV2_11/tensor_names\n",
      "save/RestoreV2_11/shape_and_slices\n",
      "save/RestoreV2_11\n",
      "save/Assign_11\n",
      "save/restore_all\n"
     ]
    }
   ],
   "source": [
    "for op in tf.get_default_graph().get_operations():\n",
    "    print(op.name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Oops, that's a lot of operations! It's much easier to use TensorBoard to visualize the graph. The following hack will allow you to visualize the graph within Jupyter (if it does not work with your browser, you will need to use a `FileWriter` to save the graph and then visualize it in TensorBoard):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from IPython.display import clear_output, Image, display, HTML\n",
    "\n",
    "def strip_consts(graph_def, max_const_size=32):\n",
    "    \"\"\"Strip large constant values from graph_def.\"\"\"\n",
    "    strip_def = tf.GraphDef()\n",
    "    for n0 in graph_def.node:\n",
    "        n = strip_def.node.add() \n",
    "        n.MergeFrom(n0)\n",
    "        if n.op == 'Const':\n",
    "            tensor = n.attr['value'].tensor\n",
    "            size = len(tensor.tensor_content)\n",
    "            if size > max_const_size:\n",
    "                tensor.tensor_content = b\"<stripped %d bytes>\"%size\n",
    "    return strip_def\n",
    "\n",
    "def show_graph(graph_def, max_const_size=32):\n",
    "    \"\"\"Visualize TensorFlow graph.\"\"\"\n",
    "    if hasattr(graph_def, 'as_graph_def'):\n",
    "        graph_def = graph_def.as_graph_def()\n",
    "    strip_def = strip_consts(graph_def, max_const_size=max_const_size)\n",
    "    code = \"\"\"\n",
    "        <script>\n",
    "          function load() {{\n",
    "            document.getElementById(\"{id}\").pbtxt = {data};\n",
    "          }}\n",
    "        </script>\n",
    "        <link rel=\"import\" href=\"https://tensorboard.appspot.com/tf-graph-basic.build.html\" onload=load()>\n",
    "        <div style=\"height:600px\">\n",
    "          <tf-graph-basic id=\"{id}\"></tf-graph-basic>\n",
    "        </div>\n",
    "    \"\"\".format(data=repr(str(strip_def)), id='graph'+str(np.random.rand()))\n",
    "\n",
    "    iframe = \"\"\"\n",
    "        <iframe seamless style=\"width:1200px;height:620px;border:0\" srcdoc=\"{}\"></iframe>\n",
    "    \"\"\".format(code.replace('\"', '&quot;'))\n",
    "    display(HTML(iframe))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "        <iframe seamless style=\"width:1200px;height:620px;border:0\" srcdoc=\"\n",
       "        <script>\n",
       "          function load() {\n",
       "            document.getElementById(&quot;graph0.3745401188473625&quot;).pbtxt = 'node {\\n  name: &quot;X&quot;\\n  op: &quot;Placeholder&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: -1\\n        }\\n        dim {\\n          size: 784\\n        }\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;y&quot;\\n  op: &quot;Placeholder&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT64\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        unknown_rank: true\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/shape&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 2\\n          }\\n        }\\n        tensor_content: &quot;\\\\020\\\\003\\\\000\\\\000,\\\\001\\\\000\\\\000&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/min&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: -0.07439795136451721\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/max&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 0.07439795136451721\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  op: &quot;RandomUniform&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/shape&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;seed&quot;\\n    value {\\n      i: 42\\n    }\\n  }\\n  attr {\\n    key: &quot;seed2&quot;\\n    value {\\n      i: 5\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/sub&quot;\\n  op: &quot;Sub&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/max&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/sub&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Initializer/random_uniform&quot;\\n  op: &quot;Add&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/mul&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 784\\n        }\\n        dim {\\n          size: 300\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  input: &quot;hidden1/kernel/Initializer/random_uniform&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden1/kernel/read&quot;\\n  input: &quot;hidden1/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum/reduction_indices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n        }\\n        int_val: 1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum&quot;\\n  op: &quot;Sum&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum/reduction_indices&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Rsqrt&quot;\\n  op: &quot;Rsqrt&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Sum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1/y&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden1/kernel/read&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1/y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv/y&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv&quot;\\n  op: &quot;RealDiv&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Const&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv/y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Minimum&quot;\\n  op: &quot;Minimum&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Rsqrt&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/truediv&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_2&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_1&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/Minimum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm/mul_2&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/kernel/Regularizer/max_norm&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  input: &quot;dnn/hidden1/kernel/Regularizer/clip_by_norm&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 300\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 300\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/bias&quot;\\n  input: &quot;hidden1/bias/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden1/bias&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;X&quot;\\n  input: &quot;hidden1/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/BiasAdd&quot;\\n  op: &quot;BiasAdd&quot;\\n  input: &quot;dnn/hidden1/MatMul&quot;\\n  input: &quot;hidden1/bias/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden1/Relu&quot;\\n  op: &quot;Relu&quot;\\n  input: &quot;dnn/hidden1/BiasAdd&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/shape&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 2\\n          }\\n        }\\n        tensor_content: &quot;,\\\\001\\\\000\\\\0002\\\\000\\\\000\\\\000&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/min&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: -0.13093073666095734\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/max&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 0.13093073666095734\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  op: &quot;RandomUniform&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/shape&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;seed&quot;\\n    value {\\n      i: 42\\n    }\\n  }\\n  attr {\\n    key: &quot;seed2&quot;\\n    value {\\n      i: 35\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/sub&quot;\\n  op: &quot;Sub&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/max&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/sub&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Initializer/random_uniform&quot;\\n  op: &quot;Add&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/mul&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 300\\n        }\\n        dim {\\n          size: 50\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  input: &quot;hidden2/kernel/Initializer/random_uniform&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden2/kernel/read&quot;\\n  input: &quot;hidden2/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum/reduction_indices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n        }\\n        int_val: 1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum&quot;\\n  op: &quot;Sum&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum/reduction_indices&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Rsqrt&quot;\\n  op: &quot;Rsqrt&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Sum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1/y&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;hidden2/kernel/read&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1/y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv/y&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv&quot;\\n  op: &quot;RealDiv&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Const&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv/y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Minimum&quot;\\n  op: &quot;Minimum&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Rsqrt&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/truediv&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_2&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_1&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/Minimum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm/mul_2&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/kernel/Regularizer/max_norm&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  input: &quot;dnn/hidden2/kernel/Regularizer/clip_by_norm&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 50\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 50\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/bias&quot;\\n  input: &quot;hidden2/bias/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden2/bias&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;dnn/hidden1/Relu&quot;\\n  input: &quot;hidden2/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/BiasAdd&quot;\\n  op: &quot;BiasAdd&quot;\\n  input: &quot;dnn/hidden2/MatMul&quot;\\n  input: &quot;hidden2/bias/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/hidden2/Relu&quot;\\n  op: &quot;Relu&quot;\\n  input: &quot;dnn/hidden2/BiasAdd&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/shape&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 2\\n          }\\n        }\\n        tensor_content: &quot;2\\\\000\\\\000\\\\000\\\\n\\\\000\\\\000\\\\000&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/min&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: -0.3162277638912201\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/max&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 0.3162277638912201\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  op: &quot;RandomUniform&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/shape&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;seed&quot;\\n    value {\\n      i: 42\\n    }\\n  }\\n  attr {\\n    key: &quot;seed2&quot;\\n    value {\\n      i: 65\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/sub&quot;\\n  op: &quot;Sub&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/max&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/RandomUniform&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/sub&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Initializer/random_uniform&quot;\\n  op: &quot;Add&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/mul&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform/min&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 50\\n        }\\n        dim {\\n          size: 10\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/kernel&quot;\\n  input: &quot;outputs/kernel/Initializer/random_uniform&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;outputs/kernel&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 10\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 10\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/bias&quot;\\n  input: &quot;outputs/bias/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;outputs/bias&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/outputs/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;dnn/hidden2/Relu&quot;\\n  input: &quot;outputs/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;dnn/outputs/BiasAdd&quot;\\n  op: &quot;BiasAdd&quot;\\n  input: &quot;dnn/outputs/MatMul&quot;\\n  input: &quot;outputs/bias/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/Shape&quot;\\n  op: &quot;Shape&quot;\\n  input: &quot;y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT64\\n    }\\n  }\\n  attr {\\n    key: &quot;out_type&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits&quot;\\n  op: &quot;SparseSoftmaxCrossEntropyWithLogits&quot;\\n  input: &quot;dnn/outputs/BiasAdd&quot;\\n  input: &quot;y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tlabels&quot;\\n    value {\\n      type: DT_INT64\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;loss/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        int_val: 0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;loss/loss&quot;\\n  op: &quot;Mean&quot;\\n  input: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits&quot;\\n  input: &quot;loss/Const&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/Shape&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n          }\\n        }\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 1.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/Fill&quot;\\n  op: &quot;Fill&quot;\\n  input: &quot;train/gradients/Shape&quot;\\n  input: &quot;train/gradients/Const&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Reshape/shape&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        int_val: 1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Reshape&quot;\\n  op: &quot;Reshape&quot;\\n  input: &quot;train/gradients/Fill&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Reshape/shape&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tshape&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Shape&quot;\\n  op: &quot;Shape&quot;\\n  input: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;out_type&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Tile&quot;\\n  op: &quot;Tile&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Reshape&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Shape&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tmultiples&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Shape_1&quot;\\n  op: &quot;Shape&quot;\\n  input: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;out_type&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Shape_2&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n          }\\n        }\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        int_val: 0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Prod&quot;\\n  op: &quot;Prod&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Shape_1&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Const&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Const_1&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        int_val: 0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Prod_1&quot;\\n  op: &quot;Prod&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Shape_2&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Const_1&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Maximum/y&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n        }\\n        int_val: 1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Maximum&quot;\\n  op: &quot;Maximum&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Prod_1&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Maximum/y&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/floordiv&quot;\\n  op: &quot;FloorDiv&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Prod&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Maximum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/loss_grad/Shape_1&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/Cast&quot;\\n  op: &quot;Cast&quot;\\n  input: &quot;train/gradients/loss/loss_grad/floordiv&quot;\\n  attr {\\n    key: &quot;DstT&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;SrcT&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/loss_grad/truediv&quot;\\n  op: &quot;RealDiv&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Tile&quot;\\n  input: &quot;train/gradients/loss/loss_grad/Cast&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/zeros_like&quot;\\n  op: &quot;ZerosLike&quot;\\n  input: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits:1&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/PreventGradient&quot;\\n  op: &quot;PreventGradient&quot;\\n  input: &quot;loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits:1&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;message&quot;\\n    value {\\n      s: &quot;Currently there is no way to take the second derivative of sparse_softmax_cross_entropy_with_logits due to the fused implementation\\\\\\'s interaction with tf.gradients()&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims/dim&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n        }\\n        int_val: -1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims&quot;\\n  op: &quot;ExpandDims&quot;\\n  input: &quot;train/gradients/loss/loss_grad/truediv&quot;\\n  input: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims/dim&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tdim&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul&quot;\\n  op: &quot;Mul&quot;\\n  input: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/ExpandDims&quot;\\n  input: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/PreventGradient&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/BiasAdd_grad/BiasAddGrad&quot;\\n  op: &quot;BiasAddGrad&quot;\\n  input: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul&quot;\\n  input: &quot;^train/gradients/dnn/outputs/BiasAdd_grad/BiasAddGrad&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul&quot;\\n  input: &quot;^train/gradients/dnn/outputs/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/loss/SparseSoftmaxCrossEntropyWithLogits/SparseSoftmaxCrossEntropyWithLogits_grad/mul&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/outputs/BiasAdd_grad/BiasAddGrad&quot;\\n  input: &quot;^train/gradients/dnn/outputs/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/outputs/BiasAdd_grad/BiasAddGrad&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/MatMul_grad/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency&quot;\\n  input: &quot;outputs/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/MatMul_grad/MatMul_1&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;dnn/hidden2/Relu&quot;\\n  input: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/MatMul_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/dnn/outputs/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/outputs/MatMul_grad/MatMul_1&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/outputs/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/outputs/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/outputs/MatMul_grad/MatMul&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/outputs/MatMul_grad/MatMul_1&quot;\\n  input: &quot;^train/gradients/dnn/outputs/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/outputs/MatMul_grad/MatMul_1&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/Relu_grad/ReluGrad&quot;\\n  op: &quot;ReluGrad&quot;\\n  input: &quot;train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency&quot;\\n  input: &quot;dnn/hidden2/Relu&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/BiasAddGrad&quot;\\n  op: &quot;BiasAddGrad&quot;\\n  input: &quot;train/gradients/dnn/hidden2/Relu_grad/ReluGrad&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/Relu_grad/ReluGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/BiasAdd_grad/BiasAddGrad&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden2/Relu_grad/ReluGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden2/Relu_grad/ReluGrad&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/BiasAddGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden2/BiasAdd_grad/BiasAddGrad&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/MatMul_grad/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency&quot;\\n  input: &quot;hidden2/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/MatMul_grad/MatMul_1&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;dnn/hidden1/Relu&quot;\\n  input: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/MatMul_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/MatMul_grad/MatMul_1&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden2/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden2/MatMul_grad/MatMul&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden2/MatMul_grad/MatMul_1&quot;\\n  input: &quot;^train/gradients/dnn/hidden2/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden2/MatMul_grad/MatMul_1&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/Relu_grad/ReluGrad&quot;\\n  op: &quot;ReluGrad&quot;\\n  input: &quot;train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency&quot;\\n  input: &quot;dnn/hidden1/Relu&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/BiasAddGrad&quot;\\n  op: &quot;BiasAddGrad&quot;\\n  input: &quot;train/gradients/dnn/hidden1/Relu_grad/ReluGrad&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;data_format&quot;\\n    value {\\n      s: &quot;NHWC&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/Relu_grad/ReluGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/BiasAdd_grad/BiasAddGrad&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden1/Relu_grad/ReluGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden1/Relu_grad/ReluGrad&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/BiasAddGrad&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/BiasAdd_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden1/BiasAdd_grad/BiasAddGrad&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/MatMul_grad/MatMul&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency&quot;\\n  input: &quot;hidden1/kernel/read&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/MatMul_grad/MatMul_1&quot;\\n  op: &quot;MatMul&quot;\\n  input: &quot;X&quot;\\n  input: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_a&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;transpose_b&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/MatMul_grad/tuple/group_deps&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/MatMul_grad/MatMul_1&quot;\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/MatMul_grad/tuple/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden1/MatMul_grad/MatMul&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden1/MatMul_grad/MatMul&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/gradients/dnn/hidden1/MatMul_grad/tuple/control_dependency_1&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;train/gradients/dnn/hidden1/MatMul_grad/MatMul_1&quot;\\n  input: &quot;^train/gradients/dnn/hidden1/MatMul_grad/tuple/group_deps&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@train/gradients/dnn/hidden1/MatMul_grad/MatMul_1&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 784\\n          }\\n          dim {\\n            size: 300\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 784\\n        }\\n        dim {\\n          size: 300\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/kernel/Momentum&quot;\\n  input: &quot;hidden1/kernel/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/kernel/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden1/kernel/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 300\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 300\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/bias/Momentum&quot;\\n  input: &quot;hidden1/bias/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden1/bias/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden1/bias/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 300\\n          }\\n          dim {\\n            size: 50\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 300\\n        }\\n        dim {\\n          size: 50\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/kernel/Momentum&quot;\\n  input: &quot;hidden2/kernel/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/kernel/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden2/kernel/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 50\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 50\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/bias/Momentum&quot;\\n  input: &quot;hidden2/bias/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;hidden2/bias/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;hidden2/bias/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 50\\n          }\\n          dim {\\n            size: 10\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 50\\n        }\\n        dim {\\n          size: 10\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/kernel/Momentum&quot;\\n  input: &quot;outputs/kernel/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/kernel/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;outputs/kernel/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Momentum/Initializer/zeros&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n          dim {\\n            size: 10\\n          }\\n        }\\n        float_val: 0.0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Momentum&quot;\\n  op: &quot;VariableV2&quot;\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;container&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;shape&quot;\\n    value {\\n      shape {\\n        dim {\\n          size: 10\\n        }\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;shared_name&quot;\\n    value {\\n      s: &quot;&quot;\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Momentum/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/bias/Momentum&quot;\\n  input: &quot;outputs/bias/Momentum/Initializer/zeros&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;outputs/bias/Momentum/read&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;outputs/bias/Momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/learning_rate&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 0.009999999776482582\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/momentum&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_FLOAT\\n        tensor_shape {\\n        }\\n        float_val: 0.8999999761581421\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_hidden1/kernel/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  input: &quot;hidden1/kernel/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/hidden1/MatMul_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_hidden1/bias/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;hidden1/bias&quot;\\n  input: &quot;hidden1/bias/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/hidden1/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_hidden2/kernel/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  input: &quot;hidden2/kernel/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/hidden2/MatMul_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_hidden2/bias/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;hidden2/bias&quot;\\n  input: &quot;hidden2/bias/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/hidden2/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_outputs/kernel/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;outputs/kernel&quot;\\n  input: &quot;outputs/kernel/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/outputs/MatMul_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum/update_outputs/bias/ApplyMomentum&quot;\\n  op: &quot;ApplyMomentum&quot;\\n  input: &quot;outputs/bias&quot;\\n  input: &quot;outputs/bias/Momentum&quot;\\n  input: &quot;train/Momentum/learning_rate&quot;\\n  input: &quot;train/gradients/dnn/outputs/BiasAdd_grad/tuple/control_dependency_1&quot;\\n  input: &quot;train/Momentum/momentum&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n  attr {\\n    key: &quot;use_nesterov&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;train/Momentum&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^train/Momentum/update_hidden1/kernel/ApplyMomentum&quot;\\n  input: &quot;^train/Momentum/update_hidden1/bias/ApplyMomentum&quot;\\n  input: &quot;^train/Momentum/update_hidden2/kernel/ApplyMomentum&quot;\\n  input: &quot;^train/Momentum/update_hidden2/bias/ApplyMomentum&quot;\\n  input: &quot;^train/Momentum/update_outputs/kernel/ApplyMomentum&quot;\\n  input: &quot;^train/Momentum/update_outputs/bias/ApplyMomentum&quot;\\n}\\nnode {\\n  name: &quot;eval/in_top_k/InTopKV2/k&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT64\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT64\\n        tensor_shape {\\n        }\\n        int64_val: 1\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;eval/in_top_k/InTopKV2&quot;\\n  op: &quot;InTopKV2&quot;\\n  input: &quot;dnn/outputs/BiasAdd&quot;\\n  input: &quot;y&quot;\\n  input: &quot;eval/in_top_k/InTopKV2/k&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_INT64\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;eval/Cast&quot;\\n  op: &quot;Cast&quot;\\n  input: &quot;eval/in_top_k/InTopKV2&quot;\\n  attr {\\n    key: &quot;DstT&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;SrcT&quot;\\n    value {\\n      type: DT_BOOL\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;eval/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_INT32\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        int_val: 0\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;eval/Mean&quot;\\n  op: &quot;Mean&quot;\\n  input: &quot;eval/Cast&quot;\\n  input: &quot;eval/Const&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;Tidx&quot;\\n    value {\\n      type: DT_INT32\\n    }\\n  }\\n  attr {\\n    key: &quot;keep_dims&quot;\\n    value {\\n      b: false\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;init&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^hidden1/kernel/Assign&quot;\\n  input: &quot;^hidden1/bias/Assign&quot;\\n  input: &quot;^hidden2/kernel/Assign&quot;\\n  input: &quot;^hidden2/bias/Assign&quot;\\n  input: &quot;^outputs/kernel/Assign&quot;\\n  input: &quot;^outputs/bias/Assign&quot;\\n  input: &quot;^hidden1/kernel/Momentum/Assign&quot;\\n  input: &quot;^hidden1/bias/Momentum/Assign&quot;\\n  input: &quot;^hidden2/kernel/Momentum/Assign&quot;\\n  input: &quot;^hidden2/bias/Momentum/Assign&quot;\\n  input: &quot;^outputs/kernel/Momentum/Assign&quot;\\n  input: &quot;^outputs/bias/Momentum/Assign&quot;\\n}\\nnode {\\n  name: &quot;save/Const&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n        }\\n        string_val: &quot;model&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/SaveV2/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 12\\n          }\\n        }\\n        string_val: &quot;hidden1/bias&quot;\\n        string_val: &quot;hidden1/bias/Momentum&quot;\\n        string_val: &quot;hidden1/kernel&quot;\\n        string_val: &quot;hidden1/kernel/Momentum&quot;\\n        string_val: &quot;hidden2/bias&quot;\\n        string_val: &quot;hidden2/bias/Momentum&quot;\\n        string_val: &quot;hidden2/kernel&quot;\\n        string_val: &quot;hidden2/kernel/Momentum&quot;\\n        string_val: &quot;outputs/bias&quot;\\n        string_val: &quot;outputs/bias/Momentum&quot;\\n        string_val: &quot;outputs/kernel&quot;\\n        string_val: &quot;outputs/kernel/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/SaveV2/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 12\\n          }\\n        }\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/SaveV2&quot;\\n  op: &quot;SaveV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/SaveV2/tensor_names&quot;\\n  input: &quot;save/SaveV2/shape_and_slices&quot;\\n  input: &quot;hidden1/bias&quot;\\n  input: &quot;hidden1/bias/Momentum&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  input: &quot;hidden1/kernel/Momentum&quot;\\n  input: &quot;hidden2/bias&quot;\\n  input: &quot;hidden2/bias/Momentum&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  input: &quot;hidden2/kernel/Momentum&quot;\\n  input: &quot;outputs/bias&quot;\\n  input: &quot;outputs/bias/Momentum&quot;\\n  input: &quot;outputs/kernel&quot;\\n  input: &quot;outputs/kernel/Momentum&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/control_dependency&quot;\\n  op: &quot;Identity&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;^save/SaveV2&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@save/Const&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden1/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2/tensor_names&quot;\\n  input: &quot;save/RestoreV2/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/bias&quot;\\n  input: &quot;save/RestoreV2&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_1/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden1/bias/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_1/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_1&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_1/tensor_names&quot;\\n  input: &quot;save/RestoreV2_1/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_1&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/bias/Momentum&quot;\\n  input: &quot;save/RestoreV2_1&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_2/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_2/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_2&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_2/tensor_names&quot;\\n  input: &quot;save/RestoreV2_2/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_2&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/kernel&quot;\\n  input: &quot;save/RestoreV2_2&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_3/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden1/kernel/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_3/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_3&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_3/tensor_names&quot;\\n  input: &quot;save/RestoreV2_3/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_3&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden1/kernel/Momentum&quot;\\n  input: &quot;save/RestoreV2_3&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden1/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_4/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden2/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_4/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_4&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_4/tensor_names&quot;\\n  input: &quot;save/RestoreV2_4/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_4&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/bias&quot;\\n  input: &quot;save/RestoreV2_4&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_5/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden2/bias/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_5/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_5&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_5/tensor_names&quot;\\n  input: &quot;save/RestoreV2_5/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_5&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/bias/Momentum&quot;\\n  input: &quot;save/RestoreV2_5&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_6/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_6/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_6&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_6/tensor_names&quot;\\n  input: &quot;save/RestoreV2_6/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_6&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/kernel&quot;\\n  input: &quot;save/RestoreV2_6&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_7/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;hidden2/kernel/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_7/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_7&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_7/tensor_names&quot;\\n  input: &quot;save/RestoreV2_7/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_7&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;hidden2/kernel/Momentum&quot;\\n  input: &quot;save/RestoreV2_7&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@hidden2/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_8/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;outputs/bias&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_8/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_8&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_8/tensor_names&quot;\\n  input: &quot;save/RestoreV2_8/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_8&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/bias&quot;\\n  input: &quot;save/RestoreV2_8&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_9/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;outputs/bias/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_9/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_9&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_9/tensor_names&quot;\\n  input: &quot;save/RestoreV2_9/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_9&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/bias/Momentum&quot;\\n  input: &quot;save/RestoreV2_9&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/bias&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_10/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;outputs/kernel&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_10/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_10&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_10/tensor_names&quot;\\n  input: &quot;save/RestoreV2_10/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_10&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/kernel&quot;\\n  input: &quot;save/RestoreV2_10&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_11/tensor_names&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;outputs/kernel/Momentum&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_11/shape_and_slices&quot;\\n  op: &quot;Const&quot;\\n  attr {\\n    key: &quot;dtype&quot;\\n    value {\\n      type: DT_STRING\\n    }\\n  }\\n  attr {\\n    key: &quot;value&quot;\\n    value {\\n      tensor {\\n        dtype: DT_STRING\\n        tensor_shape {\\n          dim {\\n            size: 1\\n          }\\n        }\\n        string_val: &quot;&quot;\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/RestoreV2_11&quot;\\n  op: &quot;RestoreV2&quot;\\n  input: &quot;save/Const&quot;\\n  input: &quot;save/RestoreV2_11/tensor_names&quot;\\n  input: &quot;save/RestoreV2_11/shape_and_slices&quot;\\n  attr {\\n    key: &quot;dtypes&quot;\\n    value {\\n      list {\\n        type: DT_FLOAT\\n      }\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/Assign_11&quot;\\n  op: &quot;Assign&quot;\\n  input: &quot;outputs/kernel/Momentum&quot;\\n  input: &quot;save/RestoreV2_11&quot;\\n  attr {\\n    key: &quot;T&quot;\\n    value {\\n      type: DT_FLOAT\\n    }\\n  }\\n  attr {\\n    key: &quot;_class&quot;\\n    value {\\n      list {\\n        s: &quot;loc:@outputs/kernel&quot;\\n      }\\n    }\\n  }\\n  attr {\\n    key: &quot;use_locking&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n  attr {\\n    key: &quot;validate_shape&quot;\\n    value {\\n      b: true\\n    }\\n  }\\n}\\nnode {\\n  name: &quot;save/restore_all&quot;\\n  op: &quot;NoOp&quot;\\n  input: &quot;^save/Assign&quot;\\n  input: &quot;^save/Assign_1&quot;\\n  input: &quot;^save/Assign_2&quot;\\n  input: &quot;^save/Assign_3&quot;\\n  input: &quot;^save/Assign_4&quot;\\n  input: &quot;^save/Assign_5&quot;\\n  input: &quot;^save/Assign_6&quot;\\n  input: &quot;^save/Assign_7&quot;\\n  input: &quot;^save/Assign_8&quot;\\n  input: &quot;^save/Assign_9&quot;\\n  input: &quot;^save/Assign_10&quot;\\n  input: &quot;^save/Assign_11&quot;\\n}\\n';\n",
       "          }\n",
       "        </script>\n",
       "        <link rel=&quot;import&quot; href=&quot;https://tensorboard.appspot.com/tf-graph-basic.build.html&quot; onload=load()>\n",
       "        <div style=&quot;height:600px&quot;>\n",
       "          <tf-graph-basic id=&quot;graph0.3745401188473625&quot;></tf-graph-basic>\n",
       "        </div>\n",
       "    \"></iframe>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_graph(tf.get_default_graph())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Once you know which operations you need, you can get a handle on them using the graph's `get_operation_by_name()` or `get_tensor_by_name()` methods:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "\"The name 'eval/accuracy:0' refers to a Tensor which does not exist. The operation, 'eval/accuracy', does not exist in the graph.\"",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                   Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-32-da429e01f50b>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0my\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_default_graph\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_tensor_by_name\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"y:0\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 4\u001b[0;31m \u001b[0maccuracy\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_default_graph\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_tensor_by_name\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"eval/accuracy:0\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mget_tensor_by_name\u001b[0;34m(self, name)\u001b[0m\n\u001b[1;32m   3205\u001b[0m       raise TypeError(\"Tensor names are strings (or similar), not %s.\" %\n\u001b[1;32m   3206\u001b[0m                       type(name).__name__)\n\u001b[0;32m-> 3207\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_graph_element\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3208\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3209\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_get_tensor_by_tf_output\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtf_output\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mas_graph_element\u001b[0;34m(self, obj, allow_tensor, allow_operation)\u001b[0m\n\u001b[1;32m   3033\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3034\u001b[0m     \u001b[0;32mwith\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_lock\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3035\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_as_graph_element_locked\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mobj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3036\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3037\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_as_graph_element_locked\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mobj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36m_as_graph_element_locked\u001b[0;34m(self, obj, allow_tensor, allow_operation)\u001b[0m\n\u001b[1;32m   3075\u001b[0m           raise KeyError(\"The name %s refers to a Tensor which does not \"\n\u001b[1;32m   3076\u001b[0m                          \u001b[0;34m\"exist. The operation, %s, does not exist in the \"\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3077\u001b[0;31m                          \"graph.\" % (repr(name), repr(op_name)))\n\u001b[0m\u001b[1;32m   3078\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3079\u001b[0m           \u001b[0;32mreturn\u001b[0m \u001b[0mop\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mout_n\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyError\u001b[0m: \"The name 'eval/accuracy:0' refers to a Tensor which does not exist. The operation, 'eval/accuracy', does not exist in the graph.\""
     ]
    }
   ],
   "source": [
    "X = tf.get_default_graph().get_tensor_by_name(\"X:0\")\n",
    "y = tf.get_default_graph().get_tensor_by_name(\"y:0\")\n",
    "\n",
    "accuracy = tf.get_default_graph().get_tensor_by_name(\"eval/accuracy:0\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "\"The name 'train/GradientDescent' refers to an Operation not in the graph.\"",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                   Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-34-70cba5c4d215>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtraining_op\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_default_graph\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mget_operation_by_name\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"train/GradientDescent\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mget_operation_by_name\u001b[0;34m(self, name)\u001b[0m\n\u001b[1;32m   3161\u001b[0m       raise TypeError(\"Operation names are strings (or similar), not %s.\" %\n\u001b[1;32m   3162\u001b[0m                       type(name).__name__)\n\u001b[0;32m-> 3163\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_graph_element\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mFalse\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3164\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3165\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_get_operation_by_name_unsafe\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36mas_graph_element\u001b[0;34m(self, obj, allow_tensor, allow_operation)\u001b[0m\n\u001b[1;32m   3033\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3034\u001b[0m     \u001b[0;32mwith\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_lock\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3035\u001b[0;31m       \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_as_graph_element_locked\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mobj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3036\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3037\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_as_graph_element_locked\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mobj\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_tensor\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mallow_operation\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda/lib/python3.5/site-packages/tensorflow/python/framework/ops.py\u001b[0m in \u001b[0;36m_as_graph_element_locked\u001b[0;34m(self, obj, allow_tensor, allow_operation)\u001b[0m\n\u001b[1;32m   3093\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_nodes_by_name\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3094\u001b[0m           raise KeyError(\"The name %s refers to an Operation not in the \"\n\u001b[0;32m-> 3095\u001b[0;31m                          \"graph.\" % repr(name))\n\u001b[0m\u001b[1;32m   3096\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_nodes_by_name\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3097\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyError\u001b[0m: \"The name 'train/GradientDescent' refers to an Operation not in the graph.\""
     ]
    }
   ],
   "source": [
    "training_op = tf.get_default_graph().get_operation_by_name(\"train/GradientDescent\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./my_model_final.ckpt\n",
      "0 Test accuracy: 0.9636\n",
      "1 Test accuracy: 0.965\n",
      "2 Test accuracy: 0.9661\n",
      "3 Test accuracy: 0.9664\n",
      "4 Test accuracy: 0.9667\n",
      "5 Test accuracy: 0.9679\n",
      "6 Test accuracy: 0.9686\n",
      "7 Test accuracy: 0.9691\n",
      "8 Test accuracy: 0.9691\n",
      "9 Test accuracy: 0.9698\n",
      "10 Test accuracy: 0.9709\n",
      "11 Test accuracy: 0.9707\n",
      "12 Test accuracy: 0.9706\n",
      "13 Test accuracy: 0.9714\n",
      "14 Test accuracy: 0.9712\n",
      "15 Test accuracy: 0.9707\n",
      "16 Test accuracy: 0.9713\n",
      "17 Test accuracy: 0.9715\n",
      "18 Test accuracy: 0.9722\n",
      "19 Test accuracy: 0.9716\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    saver.restore(sess, \"./my_model_final.ckpt\")\n",
    "\n",
    "    for epoch in range(n_epochs):\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        accuracy_val = accuracy.eval(feed_dict={X: mnist.test.images,\n",
    "                                                y: mnist.test.labels})\n",
    "        print(epoch, \"Test accuracy:\", accuracy_val)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_new_model_final.ckpt\")    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, it's possible to kick off layers and add new ones while holding onto lower layers, but I doubt we'll get there in only an hour. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Avoiding Overfitting Through Regularization\n",
    "\n",
    "#### $\\ell_1$ and $\\ell_2$ regularization\n",
    "\n",
    "Let's implement $\\ell_1$ regularization manually. First, we create the model, as usual (with just one hidden layer this time, for simplicity):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "n_inputs = 28 * 28  # MNIST\n",
    "n_hidden1 = 300\n",
    "n_outputs = 10\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")\n",
    "\n",
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.relu, name=\"hidden1\")\n",
    "    logits = tf.layers.dense(hidden1, n_outputs, name=\"outputs\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, we get a handle on the layer weights, and we compute the total loss, which is equal to the sum of the usual cross entropy loss and the $\\ell_1$ loss (i.e., the absolute values of the weights):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "W1 = tf.get_default_graph().get_tensor_by_name(\"hidden1/kernel:0\")\n",
    "W2 = tf.get_default_graph().get_tensor_by_name(\"outputs/kernel:0\")\n",
    "\n",
    "scale = 0.001 # l1 regularization hyperparameter\n",
    "\n",
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,\n",
    "                                                              logits=logits)\n",
    "    base_loss = tf.reduce_mean(xentropy, name=\"avg_xentropy\")\n",
    "    reg_losses = tf.reduce_sum(tf.abs(W1)) + tf.reduce_sum(tf.abs(W2))\n",
    "    loss = tf.add(base_loss, scale * reg_losses, name=\"loss\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"eval\"):\n",
    "    correct = tf.nn.in_top_k(logits, y, 1)\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32), name=\"accuracy\")\n",
    "\n",
    "learning_rate = 0.01\n",
    "\n",
    "with tf.name_scope(\"train\"):\n",
    "    optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "    training_op = optimizer.minimize(loss)\n",
    "\n",
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Test accuracy: 0.8347\n",
      "1 Test accuracy: 0.8713\n",
      "2 Test accuracy: 0.8832\n",
      "3 Test accuracy: 0.8907\n",
      "4 Test accuracy: 0.8959\n",
      "5 Test accuracy: 0.8978\n",
      "6 Test accuracy: 0.9015\n",
      "7 Test accuracy: 0.9035\n",
      "8 Test accuracy: 0.9033\n",
      "9 Test accuracy: 0.9064\n",
      "10 Test accuracy: 0.906\n",
      "11 Test accuracy: 0.907\n",
      "12 Test accuracy: 0.9076\n",
      "13 Test accuracy: 0.9068\n",
      "14 Test accuracy: 0.9065\n",
      "15 Test accuracy: 0.9076\n",
      "16 Test accuracy: 0.9068\n",
      "17 Test accuracy: 0.9063\n",
      "18 Test accuracy: 0.9064\n",
      "19 Test accuracy: 0.9066\n"
     ]
    }
   ],
   "source": [
    "n_epochs = 20\n",
    "batch_size = 200\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        accuracy_val = accuracy.eval(feed_dict={X: mnist.test.images,\n",
    "                                                y: mnist.test.labels})\n",
    "        print(epoch, \"Test accuracy:\", accuracy_val)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This could be prohibitively slow if we have many layers. Instead, use the `l1_regularizer()`, *etc.* functions. First, define the regularization parameters and your base loss function.\n",
    "```python\n",
    "scale = 0.001 # l1 regularization hyperparameter\n",
    "\n",
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,\n",
    "                                                              logits=logits)\n",
    "    base_loss = tf.reduce_mean(xentropy, name=\"avg_xentropy\")\n",
    "```\n",
    "Next, construct your network using the `kernel_regularizer` argument.\n",
    "```python\n",
    "my_dense_layer = partial(\n",
    "    tf.layers.dense, activation=tf.nn.relu,\n",
    "    kernel_regularizer=tf.contrib.layers.l1_regularizer(scale))\n",
    "\n",
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = my_dense_layer(X, n_hidden1, name=\"hidden1\")\n",
    "    hidden2 = my_dense_layer(hidden1, n_hidden2, name=\"hidden2\")\n",
    "    logits = my_dense_layer(hidden2, n_outputs, activation=None,\n",
    "                            name=\"outputs\")\n",
    "```\n",
    "Then, add the regularization losses to your base loss to obtain your loss function.\n",
    "```python\n",
    "reg_losses = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)\n",
    "loss = tf.add_n([base_loss] + reg_losses, name=\"loss\")\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Max norm\n",
    "\n",
    "Max-norm regularization constrains the weights $w$ of the incoming connections such that $\\|w\\|_2 \\leq r$ for some hyperparameter $r$. This is typically implemented by clipping $w$ after each training step according to $w \\leftarrow w \\cdot r / \\|w\\|_2$. To implement max-norm \"regularization\" we'll wind up getting a handle on each of the weights and creating an `Operation` that will reassign a *clipped* set of weights to those tensors. We'll have the `Session` object `run` that `Operation` on each iteration after the `training_op` call.\n",
    "\n",
    "**Pet peeve:** I would call this a *constraint* and not merely regularization since $w$ **cannot** be greater than $r$.\n",
    "\n",
    "First, build the neural network: a plain and simple neural net for MNIST with just 2 hidden layers:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "n_inputs = 28 * 28\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 50\n",
    "n_outputs = 10\n",
    "\n",
    "learning_rate = 0.01\n",
    "momentum = 0.9\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")\n",
    "\n",
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.relu, name=\"hidden1\")\n",
    "    hidden2 = tf.layers.dense(hidden1, n_hidden2, activation=tf.nn.relu, name=\"hidden2\")\n",
    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")\n",
    "\n",
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits)\n",
    "    loss = tf.reduce_mean(xentropy, name=\"loss\")\n",
    "\n",
    "with tf.name_scope(\"train\"):\n",
    "    optimizer = tf.train.MomentumOptimizer(learning_rate, momentum)\n",
    "    training_op = optimizer.minimize(loss)    \n",
    "\n",
    "with tf.name_scope(\"eval\"):\n",
    "    correct = tf.nn.in_top_k(logits, y, 1)\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, let's get a handle on the first hidden layer's weight and create an operation that will compute the clipped weights using the `clip_by_norm()` function. Then we create an assignment operation to assign the clipped weights to the weights variable:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "threshold = 1.0\n",
    "weights = tf.get_default_graph().get_tensor_by_name(\"hidden1/kernel:0\")\n",
    "clipped_weights = tf.clip_by_norm(weights, clip_norm=threshold, axes=1)\n",
    "clip_weights = tf.assign(weights, clipped_weights)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can do this as well for the second hidden layer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "weights2 = tf.get_default_graph().get_tensor_by_name(\"hidden2/kernel:0\")\n",
    "clipped_weights2 = tf.clip_by_norm(weights2, clip_norm=threshold, axes=1)\n",
    "clip_weights2 = tf.assign(weights2, clipped_weights2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's add an initializer and a saver:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And now we can train the model. It's pretty much as usual, except that right after running the `training_op`, we run the `clip_weights` and `clip_weights2` operations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_epochs = 20\n",
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Test accuracy: 0.9525\n",
      "1 Test accuracy: 0.9688\n",
      "2 Test accuracy: 0.9732\n",
      "3 Test accuracy: 0.9755\n",
      "4 Test accuracy: 0.975\n",
      "5 Test accuracy: 0.9769\n",
      "6 Test accuracy: 0.9795\n",
      "7 Test accuracy: 0.9778\n",
      "8 Test accuracy: 0.979\n",
      "9 Test accuracy: 0.9792\n",
      "10 Test accuracy: 0.9811\n",
      "11 Test accuracy: 0.9802\n",
      "12 Test accuracy: 0.9826\n",
      "13 Test accuracy: 0.9816\n",
      "14 Test accuracy: 0.9824\n",
      "15 Test accuracy: 0.9817\n",
      "16 Test accuracy: 0.9823\n",
      "17 Test accuracy: 0.9827\n",
      "18 Test accuracy: 0.9829\n",
      "19 Test accuracy: 0.9827\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:                                              # not shown in the book\n",
    "    init.run()                                                          # not shown\n",
    "    for epoch in range(n_epochs):                                       # not shown\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):  # not shown\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)       # not shown\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "            clip_weights.eval()\n",
    "            clip_weights2.eval()                                        # not shown\n",
    "        acc_test = accuracy.eval(feed_dict={X: mnist.test.images,       # not shown\n",
    "                                            y: mnist.test.labels})      # not shown\n",
    "        print(epoch, \"Test accuracy:\", acc_test)                        # not shown\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")               # not shown"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The implementation above is straightforward and it works fine, but it is a bit messy. A better approach is to define a `max_norm_regularizer()` function:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def max_norm_regularizer(threshold, axes=1, name=\"max_norm\",\n",
    "                         collection=\"max_norm\"):\n",
    "    def max_norm(weights):\n",
    "        clipped = tf.clip_by_norm(weights, clip_norm=threshold, axes=axes)\n",
    "        clip_weights = tf.assign(weights, clipped, name=name)\n",
    "        tf.add_to_collection(collection, clip_weights)\n",
    "        return None # there is no regularization loss term\n",
    "    return max_norm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then you can call this function to get a max norm regularizer (with the threshold you want). When you create a hidden layer, you can pass this regularizer to the `kernel_regularizer` argument:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "n_inputs = 28 * 28\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 50\n",
    "n_outputs = 10\n",
    "\n",
    "learning_rate = 0.01\n",
    "momentum = 0.9\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int64, shape=(None), name=\"y\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "max_norm_reg = max_norm_regularizer(threshold=1.0)\n",
    "\n",
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, activation=tf.nn.relu,\n",
    "                              kernel_regularizer=max_norm_reg, name=\"hidden1\")\n",
    "    hidden2 = tf.layers.dense(hidden1, n_hidden2, activation=tf.nn.relu,\n",
    "                              kernel_regularizer=max_norm_reg, name=\"hidden2\")\n",
    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits)\n",
    "    loss = tf.reduce_mean(xentropy, name=\"loss\")\n",
    "\n",
    "with tf.name_scope(\"train\"):\n",
    "    optimizer = tf.train.MomentumOptimizer(learning_rate, momentum)\n",
    "    training_op = optimizer.minimize(loss)    \n",
    "\n",
    "with tf.name_scope(\"eval\"):\n",
    "    correct = tf.nn.in_top_k(logits, y, 1)\n",
    "    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))\n",
    "\n",
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Training is as usual, except you must run the weights clipping operations after each training operation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n_epochs = 20\n",
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Test accuracy: 0.9534\n",
      "1 Test accuracy: 0.9657\n",
      "2 Test accuracy: 0.9729\n",
      "3 Test accuracy: 0.9718\n",
      "4 Test accuracy: 0.9752\n",
      "5 Test accuracy: 0.9749\n",
      "6 Test accuracy: 0.9777\n",
      "7 Test accuracy: 0.9796\n",
      "8 Test accuracy: 0.9791\n",
      "9 Test accuracy: 0.9801\n",
      "10 Test accuracy: 0.9792\n",
      "11 Test accuracy: 0.9801\n",
      "12 Test accuracy: 0.9811\n",
      "13 Test accuracy: 0.9815\n",
      "14 Test accuracy: 0.9802\n",
      "15 Test accuracy: 0.982\n",
      "16 Test accuracy: 0.9805\n",
      "17 Test accuracy: 0.9813\n",
      "18 Test accuracy: 0.9813\n",
      "19 Test accuracy: 0.9812\n"
     ]
    }
   ],
   "source": [
    "clip_all_weights = tf.get_collection(\"max_norm\")\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for iteration in range(mnist.train.num_examples // batch_size):\n",
    "            X_batch, y_batch = mnist.train.next_batch(batch_size)\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "            sess.run(clip_all_weights)\n",
    "        acc_test = accuracy.eval(feed_dict={X: mnist.test.images,     # not shown in the book\n",
    "                                            y: mnist.test.labels})    # not shown\n",
    "        print(epoch, \"Test accuracy:\", acc_test)                      # not shown\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")             # not shown"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Faster Optimizers\n",
    "\n",
    "* Momentum Optimization\n",
    "\n",
    "\\begin{align*}\n",
    "m &\\leftarrow \\beta m - \\eta \\nabla_\\theta J(\\theta)\n",
    "\\\\\n",
    "\\theta &\\leftarrow \\theta + m\n",
    "\\end{align*}\n",
    "\n",
    "* Nesterov Accelerated Gradient\n",
    "\n",
    "\\begin{align*}\n",
    "m &\\leftarrow \\beta m - \\eta \\nabla_\\theta J(\\theta + \\beta m)\n",
    "\\\\\n",
    "\\theta & \\leftarrow \\theta + m\n",
    "\\end{align*}\n",
    "\n",
    "* Adagrad \n",
    "* RMSProp\n",
    "* Adam\n",
    "* Learning Rate Scheduling / Annealing\n",
    "\n",
    "For a complete list, see [the documentation](https://www.tensorflow.org/api_guides/python/train).\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.4"
  },
  "toc": {
   "colors": {
    "hover_highlight": "#DAA520",
    "navigate_num": "#000000",
    "navigate_text": "#333333",
    "running_highlight": "#FF0000",
    "selected_highlight": "#FFD700",
    "sidebar_border": "#EEEEEE",
    "wrapper_background": "#FFFFFF"
   },
   "moveMenuLeft": true,
   "nav_menu": {
    "height": "94px",
    "width": "294px"
   },
   "navigate_menu": true,
   "number_sections": false,
   "sideBar": true,
   "threshold": 4,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false,
   "widenNotebook": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}