{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Chapter 10 – Introduction to Artificial Neural Networks**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "_This notebook contains all the sample code and solutions to the exercises in chapter 10._"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Setup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's make sure this notebook works well in both python 2 and 3, import a few common modules, ensure MatplotLib plots figures inline and prepare a function to save the figures:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# To support both python 2 and python 3\n",
    "from __future__ import division, print_function, unicode_literals\n",
    "\n",
    "# Common imports\n",
    "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\n",
    "\n",
    "# Where to save the figures\n",
    "PROJECT_ROOT_DIR = \".\"\n",
    "CHAPTER_ID = \"ann\"\n",
    "\n",
    "def save_fig(fig_id, tight_layout=True):\n",
    "    path = os.path.join(PROJECT_ROOT_DIR, \"images\", CHAPTER_ID, fig_id + \".png\")\n",
    "    print(\"Saving figure\", fig_id)\n",
    "    if tight_layout:\n",
    "        plt.tight_layout()\n",
    "    plt.savefig(path, format='png', dpi=300)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Perceptrons"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from sklearn.datasets import load_iris\n",
    "from sklearn.linear_model import Perceptron\n",
    "\n",
    "iris = load_iris()\n",
    "X = iris.data[:, (2, 3)]  # petal length, petal width\n",
    "y = (iris.target == 0).astype(np.int)\n",
    "\n",
    "per_clf = Perceptron(max_iter=100, random_state=42)\n",
    "per_clf.fit(X, y)\n",
    "\n",
    "y_pred = per_clf.predict([[2, 0.5]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "a = -per_clf.coef_[0][0] / per_clf.coef_[0][1]\n",
    "b = -per_clf.intercept_ / per_clf.coef_[0][1]\n",
    "\n",
    "axes = [0, 5, 0, 2]\n",
    "\n",
    "x0, x1 = np.meshgrid(\n",
    "        np.linspace(axes[0], axes[1], 500).reshape(-1, 1),\n",
    "        np.linspace(axes[2], axes[3], 200).reshape(-1, 1),\n",
    "    )\n",
    "X_new = np.c_[x0.ravel(), x1.ravel()]\n",
    "y_predict = per_clf.predict(X_new)\n",
    "zz = y_predict.reshape(x0.shape)\n",
    "\n",
    "plt.figure(figsize=(10, 4))\n",
    "plt.plot(X[y==0, 0], X[y==0, 1], \"bs\", label=\"Not Iris-Setosa\")\n",
    "plt.plot(X[y==1, 0], X[y==1, 1], \"yo\", label=\"Iris-Setosa\")\n",
    "\n",
    "plt.plot([axes[0], axes[1]], [a * axes[0] + b, a * axes[1] + b], \"k-\", linewidth=3)\n",
    "from matplotlib.colors import ListedColormap\n",
    "custom_cmap = ListedColormap(['#9898ff', '#fafab0'])\n",
    "\n",
    "plt.contourf(x0, x1, zz, cmap=custom_cmap)\n",
    "plt.xlabel(\"Petal length\", fontsize=14)\n",
    "plt.ylabel(\"Petal width\", fontsize=14)\n",
    "plt.legend(loc=\"lower right\", fontsize=14)\n",
    "plt.axis(axes)\n",
    "\n",
    "save_fig(\"perceptron_iris_plot\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Activation functions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def logit(z):\n",
    "    return 1 / (1 + np.exp(-z))\n",
    "\n",
    "def relu(z):\n",
    "    return np.maximum(0, z)\n",
    "\n",
    "def derivative(f, z, eps=0.000001):\n",
    "    return (f(z + eps) - f(z - eps))/(2 * eps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "z = np.linspace(-5, 5, 200)\n",
    "\n",
    "plt.figure(figsize=(11,4))\n",
    "\n",
    "plt.subplot(121)\n",
    "plt.plot(z, np.sign(z), \"r-\", linewidth=2, label=\"Step\")\n",
    "plt.plot(z, logit(z), \"g--\", linewidth=2, label=\"Logit\")\n",
    "plt.plot(z, np.tanh(z), \"b-\", linewidth=2, label=\"Tanh\")\n",
    "plt.plot(z, relu(z), \"m-.\", linewidth=2, label=\"ReLU\")\n",
    "plt.grid(True)\n",
    "plt.legend(loc=\"center right\", fontsize=14)\n",
    "plt.title(\"Activation functions\", fontsize=14)\n",
    "plt.axis([-5, 5, -1.2, 1.2])\n",
    "\n",
    "plt.subplot(122)\n",
    "plt.plot(z, derivative(np.sign, z), \"r-\", linewidth=2, label=\"Step\")\n",
    "plt.plot(0, 0, \"ro\", markersize=5)\n",
    "plt.plot(0, 0, \"rx\", markersize=10)\n",
    "plt.plot(z, derivative(logit, z), \"g--\", linewidth=2, label=\"Logit\")\n",
    "plt.plot(z, derivative(np.tanh, z), \"b-\", linewidth=2, label=\"Tanh\")\n",
    "plt.plot(z, derivative(relu, z), \"m-.\", linewidth=2, label=\"ReLU\")\n",
    "plt.grid(True)\n",
    "#plt.legend(loc=\"center right\", fontsize=14)\n",
    "plt.title(\"Derivatives\", fontsize=14)\n",
    "plt.axis([-5, 5, -0.2, 1.2])\n",
    "\n",
    "save_fig(\"activation_functions_plot\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def heaviside(z):\n",
    "    return (z >= 0).astype(z.dtype)\n",
    "\n",
    "def sigmoid(z):\n",
    "    return 1/(1+np.exp(-z))\n",
    "\n",
    "def mlp_xor(x1, x2, activation=heaviside):\n",
    "    return activation(-activation(x1 + x2 - 1.5) + activation(x1 + x2 - 0.5) - 0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "x1s = np.linspace(-0.2, 1.2, 100)\n",
    "x2s = np.linspace(-0.2, 1.2, 100)\n",
    "x1, x2 = np.meshgrid(x1s, x2s)\n",
    "\n",
    "z1 = mlp_xor(x1, x2, activation=heaviside)\n",
    "z2 = mlp_xor(x1, x2, activation=sigmoid)\n",
    "\n",
    "plt.figure(figsize=(10,4))\n",
    "\n",
    "plt.subplot(121)\n",
    "plt.contourf(x1, x2, z1)\n",
    "plt.plot([0, 1], [0, 1], \"gs\", markersize=20)\n",
    "plt.plot([0, 1], [1, 0], \"y^\", markersize=20)\n",
    "plt.title(\"Activation function: heaviside\", fontsize=14)\n",
    "plt.grid(True)\n",
    "\n",
    "plt.subplot(122)\n",
    "plt.contourf(x1, x2, z2)\n",
    "plt.plot([0, 1], [0, 1], \"gs\", markersize=20)\n",
    "plt.plot([0, 1], [1, 0], \"y^\", markersize=20)\n",
    "plt.title(\"Activation function: sigmoid\", fontsize=14)\n",
    "plt.grid(True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# FNN for MNIST"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Using the Estimator API (formerly `tf.contrib.learn`)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Warning**: `tf.examples.tutorials.mnist` is deprecated. We will use `tf.keras.datasets.mnist` instead. Moreover, the `tf.contrib.learn` API was promoted to `tf.estimators` and `tf.feature_columns`, and it has changed considerably. In particular, there is no `infer_real_valued_columns_from_input()` function or `SKCompat` class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()\n",
    "X_train = X_train.astype(np.float32).reshape(-1, 28*28) / 255.0\n",
    "X_test = X_test.astype(np.float32).reshape(-1, 28*28) / 255.0\n",
    "y_train = y_train.astype(np.int32)\n",
    "y_test = y_test.astype(np.int32)\n",
    "X_valid, X_train = X_train[:5000], X_train[5000:]\n",
    "y_valid, y_train = y_train[:5000], y_train[5000:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "feature_cols = [tf.feature_column.numeric_column(\"X\", shape=[28 * 28])]\n",
    "dnn_clf = tf.estimator.DNNClassifier(hidden_units=[300,100], n_classes=10,\n",
    "                                     feature_columns=feature_cols)\n",
    "\n",
    "input_fn = tf.estimator.inputs.numpy_input_fn(\n",
    "    x={\"X\": X_train}, y=y_train, num_epochs=40, batch_size=50, shuffle=True)\n",
    "dnn_clf.train(input_fn=input_fn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
    "    x={\"X\": X_test}, y=y_test, shuffle=False)\n",
    "eval_results = dnn_clf.evaluate(input_fn=test_input_fn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred_iter = dnn_clf.predict(input_fn=test_input_fn)\n",
    "y_pred = list(y_pred_iter)\n",
    "y_pred[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Using plain TensorFlow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "n_inputs = 28*28  # MNIST\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 100\n",
    "n_outputs = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def neuron_layer(X, n_neurons, name, activation=None):\n",
    "    with tf.name_scope(name):\n",
    "        n_inputs = int(X.get_shape()[1])\n",
    "        stddev = 2 / np.sqrt(n_inputs)\n",
    "        init = tf.truncated_normal((n_inputs, n_neurons), stddev=stddev)\n",
    "        W = tf.Variable(init, name=\"kernel\")\n",
    "        b = tf.Variable(tf.zeros([n_neurons]), name=\"bias\")\n",
    "        Z = tf.matmul(X, W) + b\n",
    "        if activation is not None:\n",
    "            return activation(Z)\n",
    "        else:\n",
    "            return Z"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = neuron_layer(X, n_hidden1, name=\"hidden1\",\n",
    "                           activation=tf.nn.relu)\n",
    "    hidden2 = neuron_layer(hidden1, n_hidden2, name=\"hidden2\",\n",
    "                           activation=tf.nn.relu)\n",
    "    logits = neuron_layer(hidden2, n_outputs, name=\"outputs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.name_scope(\"loss\"):\n",
    "    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y,\n",
    "                                                              logits=logits)\n",
    "    loss = tf.reduce_mean(xentropy, name=\"loss\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "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": 21,
   "metadata": {},
   "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))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_epochs = 40\n",
    "batch_size = 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def shuffle_batch(X, y, batch_size):\n",
    "    rnd_idx = np.random.permutation(len(X))\n",
    "    n_batches = len(X) // batch_size\n",
    "    for batch_idx in np.array_split(rnd_idx, n_batches):\n",
    "        X_batch, y_batch = X[batch_idx], y[batch_idx]\n",
    "        yield X_batch, y_batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        acc_batch = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
    "        acc_val = accuracy.eval(feed_dict={X: X_valid, y: y_valid})\n",
    "        print(epoch, \"Batch accuracy:\", acc_batch, \"Val accuracy:\", acc_val)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.Session() as sess:\n",
    "    saver.restore(sess, \"./my_model_final.ckpt\") # or better, use save_path\n",
    "    X_new_scaled = X_test[:20]\n",
    "    Z = logits.eval(feed_dict={X: X_new_scaled})\n",
    "    y_pred = np.argmax(Z, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(\"Predicted classes:\", y_pred)\n",
    "print(\"Actual classes:   \", y_test[:20])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow_graph_in_jupyter import show_graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_graph(tf.get_default_graph())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Using `dense()` instead of `neuron_layer()`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note: previous releases of the book used `tensorflow.contrib.layers.fully_connected()` rather than `tf.layers.dense()` (which did not exist when this chapter was written). It is now preferable to use `tf.layers.dense()`, because anything in the contrib module may change or be deleted without notice. The `dense()` function is almost identical to the `fully_connected()` function, except for a few minor differences:\n",
    "* several parameters are renamed: `scope` becomes `name`, `activation_fn` becomes `activation` (and similarly the `_fn` suffix is removed from other parameters such as `normalizer_fn`), `weights_initializer` becomes `kernel_initializer`, etc.\n",
    "* the default `activation` is now `None` rather than `tf.nn.relu`.\n",
    "* a few more differences are presented in chapter 11."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_inputs = 28*28  # MNIST\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 100\n",
    "n_outputs = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\") "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, name=\"hidden1\",\n",
    "                              activation=tf.nn.relu)\n",
    "    hidden2 = tf.layers.dense(hidden1, n_hidden2, name=\"hidden2\",\n",
    "                              activation=tf.nn.relu)\n",
    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")\n",
    "    y_proba = tf.nn.softmax(logits)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "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": 34,
   "metadata": {},
   "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": 35,
   "metadata": {},
   "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))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_epochs = 20\n",
    "n_batches = 50\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    init.run()\n",
    "    for epoch in range(n_epochs):\n",
    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        acc_batch = accuracy.eval(feed_dict={X: X_batch, y: y_batch})\n",
    "        acc_valid = accuracy.eval(feed_dict={X: X_valid, y: y_valid})\n",
    "        print(epoch, \"Batch accuracy:\", acc_batch, \"Validation accuracy:\", acc_valid)\n",
    "\n",
    "    save_path = saver.save(sess, \"./my_model_final.ckpt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_graph(tf.get_default_graph())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Exercise solutions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. to 8."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "See appendix A."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 9."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "_Train a deep MLP on the MNIST dataset and see if you can get over 98% precision. Just like in the last exercise of chapter 9, try adding all the bells and whistles (i.e., save checkpoints, restore the last checkpoint in case of an interruption, add summaries, plot learning curves using TensorBoard, and so on)._"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First let's create the deep net. It's exactly the same as earlier, with just one addition: we add a `tf.summary.scalar()` to track the loss and the accuracy during training, so we can view nice learning curves using TensorBoard."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_inputs = 28*28  # MNIST\n",
    "n_hidden1 = 300\n",
    "n_hidden2 = 100\n",
    "n_outputs = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "reset_graph()\n",
    "\n",
    "X = tf.placeholder(tf.float32, shape=(None, n_inputs), name=\"X\")\n",
    "y = tf.placeholder(tf.int32, shape=(None), name=\"y\") "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.name_scope(\"dnn\"):\n",
    "    hidden1 = tf.layers.dense(X, n_hidden1, name=\"hidden1\",\n",
    "                              activation=tf.nn.relu)\n",
    "    hidden2 = tf.layers.dense(hidden1, n_hidden2, name=\"hidden2\",\n",
    "                              activation=tf.nn.relu)\n",
    "    logits = tf.layers.dense(hidden2, n_outputs, name=\"outputs\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "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",
    "    loss_summary = tf.summary.scalar('log_loss', loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "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": 45,
   "metadata": {},
   "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))\n",
    "    accuracy_summary = tf.summary.scalar('accuracy', accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we need to define the directory to write the TensorBoard logs to:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datetime import datetime\n",
    "\n",
    "def log_dir(prefix=\"\"):\n",
    "    now = datetime.utcnow().strftime(\"%Y%m%d%H%M%S\")\n",
    "    root_logdir = \"tf_logs\"\n",
    "    if prefix:\n",
    "        prefix += \"-\"\n",
    "    name = prefix + \"run-\" + now\n",
    "    return \"{}/{}/\".format(root_logdir, name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "logdir = log_dir(\"mnist_dnn\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we can create the `FileWriter` that we will use to write the TensorBoard logs:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "file_writer = tf.summary.FileWriter(logdir, tf.get_default_graph())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Hey! Why don't we implement early stopping? For this, we are going to need to use the validation set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "m, n = X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_epochs = 10001\n",
    "batch_size = 50\n",
    "n_batches = int(np.ceil(m / batch_size))\n",
    "\n",
    "checkpoint_path = \"/tmp/my_deep_mnist_model.ckpt\"\n",
    "checkpoint_epoch_path = checkpoint_path + \".epoch\"\n",
    "final_model_path = \"./my_deep_mnist_model\"\n",
    "\n",
    "best_loss = np.infty\n",
    "epochs_without_progress = 0\n",
    "max_epochs_without_progress = 50\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    if os.path.isfile(checkpoint_epoch_path):\n",
    "        # if the checkpoint file exists, restore the model and load the epoch number\n",
    "        with open(checkpoint_epoch_path, \"rb\") as f:\n",
    "            start_epoch = int(f.read())\n",
    "        print(\"Training was interrupted. Continuing at epoch\", start_epoch)\n",
    "        saver.restore(sess, checkpoint_path)\n",
    "    else:\n",
    "        start_epoch = 0\n",
    "        sess.run(init)\n",
    "\n",
    "    for epoch in range(start_epoch, n_epochs):\n",
    "        for X_batch, y_batch in shuffle_batch(X_train, y_train, batch_size):\n",
    "            sess.run(training_op, feed_dict={X: X_batch, y: y_batch})\n",
    "        accuracy_val, loss_val, accuracy_summary_str, loss_summary_str = sess.run([accuracy, loss, accuracy_summary, loss_summary], feed_dict={X: X_valid, y: y_valid})\n",
    "        file_writer.add_summary(accuracy_summary_str, epoch)\n",
    "        file_writer.add_summary(loss_summary_str, epoch)\n",
    "        if epoch % 5 == 0:\n",
    "            print(\"Epoch:\", epoch,\n",
    "                  \"\\tValidation accuracy: {:.3f}%\".format(accuracy_val * 100),\n",
    "                  \"\\tLoss: {:.5f}\".format(loss_val))\n",
    "            saver.save(sess, checkpoint_path)\n",
    "            with open(checkpoint_epoch_path, \"wb\") as f:\n",
    "                f.write(b\"%d\" % (epoch + 1))\n",
    "            if loss_val < best_loss:\n",
    "                saver.save(sess, final_model_path)\n",
    "                best_loss = loss_val\n",
    "            else:\n",
    "                epochs_without_progress += 5\n",
    "                if epochs_without_progress > max_epochs_without_progress:\n",
    "                    print(\"Early stopping\")\n",
    "                    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.remove(checkpoint_epoch_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "with tf.Session() as sess:\n",
    "    saver.restore(sess, final_model_path)\n",
    "    accuracy_val = accuracy.eval(feed_dict={X: X_test, y: y_test})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "accuracy_val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  },
  "nav_menu": {
   "height": "264px",
   "width": "369px"
  },
  "toc": {
   "navigate_menu": true,
   "number_sections": true,
   "sideBar": true,
   "threshold": 6,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}