ch13/ch13_part2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "*Python Machine Learning 3rd Edition* by [Sebastian Raschka](https://sebastianraschka.com) & [Vahid Mirjalili](http://vahidmirjalili.com), Packt Publishing Ltd. 2019\n",
    "\n",
    "Code Repository: https://github.com/rasbt/python-machine-learning-book-3rd-edition\n",
    "\n",
    "Code License: [MIT License](https://github.com/rasbt/python-machine-learning-book-3rd-edition/blob/master/LICENSE.txt)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Chapter 13: Parallelizing Neural Network Training with TensorFlow  (Part 2/2)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note that the optional watermark extension is a small IPython notebook plugin that I developed to make the code reproducible. You can just skip the following line(s)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sebastian Raschka & Vahid Mirjalili \n",
      "last updated: 2019-11-24 \n",
      "\n",
      "numpy 1.17.3\n",
      "scipy 1.3.2\n",
      "matplotlib 3.1.1\n",
      "tensorflow 2.0.0\n"
     ]
    }
   ],
   "source": [
    "%load_ext watermark\n",
    "%watermark -a \"Sebastian Raschka & Vahid Mirjalili\" -u -d -p numpy,scipy,matplotlib,tensorflow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building a neural network model in TensorFlow"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### The TensorFlow Keras API (tf.keras)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Building a linear regression model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X_train = np.arange(10).reshape((10, 1))\n",
    "y_train = np.array([1.0, 1.3, 3.1,\n",
    "                    2.0, 5.0, 6.3,\n",
    "                    6.6, 7.4, 8.0,\n",
    "                    9.0])\n",
    "\n",
    "\n",
    "plt.plot(X_train, y_train, 'o', markersize=10)\n",
    "plt.xlabel('x')\n",
    "plt.ylabel('y')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train_norm = (X_train - np.mean(X_train))/np.std(X_train)\n",
    "\n",
    "ds_train_orig = tf.data.Dataset.from_tensor_slices(\n",
    "    (tf.cast(X_train_norm, tf.float32),\n",
    "     tf.cast(y_train, tf.float32)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"my_model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "Total params: 2\n",
      "Trainable params: 2\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "class MyModel(tf.keras.Model):\n",
    "    def __init__(self):\n",
    "        super(MyModel, self).__init__()\n",
    "        self.w = tf.Variable(0.0, name='weight')\n",
    "        self.b = tf.Variable(0.0, name='bias')\n",
    "\n",
    "    def call(self, x):\n",
    "        return self.w*x + self.b\n",
    "\n",
    "\n",
    "model = MyModel()\n",
    "\n",
    "model.build(input_shape=(None, 1))\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=33, shape=(), dtype=float32, numpy=0.25>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def loss_fn(y_true, y_pred):\n",
    "    return tf.reduce_mean(tf.square(y_true - y_pred))\n",
    "\n",
    "\n",
    "## testing the function:\n",
    "yt = tf.convert_to_tensor([1.0])\n",
    "yp = tf.convert_to_tensor([1.5])\n",
    "\n",
    "loss_fn(yt, yp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(model, inputs, outputs, learning_rate):\n",
    "    with tf.GradientTape() as tape:\n",
    "        current_loss = loss_fn(model(inputs), outputs)\n",
    "    dW, db = tape.gradient(current_loss, [model.w, model.b])\n",
    "    model.w.assign_sub(learning_rate * dW)\n",
    "    model.b.assign_sub(learning_rate * db)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch    0 Step  0 Loss 43.5600\n",
      "Epoch   10 Step 100 Loss 0.7530\n",
      "Epoch   20 Step 200 Loss 20.1759\n",
      "Epoch   30 Step 300 Loss 23.3976\n",
      "Epoch   40 Step 400 Loss 6.3481\n",
      "Epoch   50 Step 500 Loss 4.6356\n",
      "Epoch   60 Step 600 Loss 0.2411\n",
      "Epoch   70 Step 700 Loss 0.2036\n",
      "Epoch   80 Step 800 Loss 3.8177\n",
      "Epoch   90 Step 900 Loss 0.9416\n",
      "Epoch  100 Step 1000 Loss 0.7035\n",
      "Epoch  110 Step 1100 Loss 0.0348\n",
      "Epoch  120 Step 1200 Loss 0.5404\n",
      "Epoch  130 Step 1300 Loss 0.1170\n",
      "Epoch  140 Step 1400 Loss 0.1195\n",
      "Epoch  150 Step 1500 Loss 0.0944\n",
      "Epoch  160 Step 1600 Loss 0.4670\n",
      "Epoch  170 Step 1700 Loss 2.0695\n",
      "Epoch  180 Step 1800 Loss 0.0020\n",
      "Epoch  190 Step 1900 Loss 0.3612\n"
     ]
    }
   ],
   "source": [
    "tf.random.set_seed(1)\n",
    "\n",
    "num_epochs = 200\n",
    "log_steps = 100\n",
    "learning_rate = 0.001\n",
    "batch_size = 1\n",
    "steps_per_epoch = int(np.ceil(len(y_train) / batch_size))\n",
    "\n",
    "\n",
    "ds_train = ds_train_orig.shuffle(buffer_size=len(y_train))\n",
    "ds_train = ds_train.repeat(count=None)\n",
    "ds_train = ds_train.batch(1)\n",
    "\n",
    "Ws, bs = [], []\n",
    "\n",
    "for i, batch in enumerate(ds_train):\n",
    "    if i >= steps_per_epoch * num_epochs:\n",
    "        break\n",
    "    Ws.append(model.w.numpy())\n",
    "    bs.append(model.b.numpy())\n",
    "\n",
    "    bx, by = batch\n",
    "    loss_val = loss_fn(model(bx), by)\n",
    "\n",
    "    train(model, bx, by, learning_rate=learning_rate)\n",
    "    if i%log_steps==0:\n",
    "        print('Epoch {:4d} Step {:2d} Loss {:6.4f}'.format(\n",
    "              int(i/steps_per_epoch), i, loss_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Final Parameters: 2.6576622 4.8798566\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 936x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print('Final Parameters:', model.w.numpy(), model.b.numpy())\n",
    "\n",
    "\n",
    "X_test = np.linspace(0, 9, num=100).reshape(-1, 1)\n",
    "X_test_norm = (X_test - np.mean(X_train)) / np.std(X_train)\n",
    "\n",
    "y_pred = model(tf.cast(X_test_norm, dtype=tf.float32))\n",
    "\n",
    "\n",
    "fig = plt.figure(figsize=(13, 5))\n",
    "ax = fig.add_subplot(1, 2, 1)\n",
    "plt.plot(X_train_norm, y_train, 'o', markersize=10)\n",
    "plt.plot(X_test_norm, y_pred, '--', lw=3)\n",
    "plt.legend(['Training examples', 'Linear Reg.'], fontsize=15)\n",
    "ax.set_xlabel('x', size=15)\n",
    "ax.set_ylabel('y', size=15)\n",
    "ax.tick_params(axis='both', which='major', labelsize=15)\n",
    "\n",
    "ax = fig.add_subplot(1, 2, 2)\n",
    "plt.plot(Ws, lw=3)\n",
    "plt.plot(bs, lw=3)\n",
    "plt.legend(['Weight w', 'Bias unit b'], fontsize=15)\n",
    "ax.set_xlabel('Iteration', size=15)\n",
    "ax.set_ylabel('Value', size=15)\n",
    "ax.tick_params(axis='both', which='major', labelsize=15)\n",
    "#plt.savefig('ch13-linreg-1.pdf')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model training via the .compile() and .fit() methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf.random.set_seed(1)\n",
    "model = MyModel()\n",
    "#model.build((None, 1))\n",
    "\n",
    "model.compile(optimizer='sgd', \n",
    "              loss=loss_fn,\n",
    "              metrics=['mae', 'mse'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 10 samples\n",
      "Epoch 1/200\n",
      "10/10 [==============================] - 0s 31ms/sample - loss: 27.8562 - mae: 4.5967 - mse: 27.8562\n",
      "Epoch 2/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 18.6235 - mae: 3.7249 - mse: 18.6235\n",
      "Epoch 3/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 12.5081 - mae: 3.0572 - mse: 12.5081\n",
      "Epoch 4/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 8.4484 - mae: 2.4816 - mse: 8.4484\n",
      "Epoch 5/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 5.7520 - mae: 2.0644 - mse: 5.7520\n",
      "Epoch 6/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 3.9580 - mae: 1.7283 - mse: 3.9580\n",
      "Epoch 7/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 2.7617 - mae: 1.4792 - mse: 2.7617\n",
      "Epoch 8/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 1.9714 - mae: 1.2577 - mse: 1.9714\n",
      "Epoch 9/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 1.4485 - mae: 1.0911 - mse: 1.4485\n",
      "Epoch 10/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 1.1002 - mae: 0.9636 - mse: 1.1002\n",
      "Epoch 11/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.8714 - mae: 0.8620 - mse: 0.8714\n",
      "Epoch 12/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.7190 - mae: 0.7764 - mse: 0.7190\n",
      "Epoch 13/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.6173 - mae: 0.7089 - mse: 0.6173\n",
      "Epoch 14/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.5492 - mae: 0.6419 - mse: 0.5492\n",
      "Epoch 15/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.5042 - mae: 0.6007 - mse: 0.5042\n",
      "Epoch 16/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4740 - mae: 0.5553 - mse: 0.4740\n",
      "Epoch 17/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4546 - mae: 0.5334 - mse: 0.4546\n",
      "Epoch 18/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4402 - mae: 0.5155 - mse: 0.4402  \n",
      "Epoch 19/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4310 - mae: 0.5022 - mse: 0.4310\n",
      "Epoch 20/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4248 - mae: 0.5025 - mse: 0.4248\n",
      "Epoch 21/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4211 - mae: 0.4927 - mse: 0.4211\n",
      "Epoch 22/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4175 - mae: 0.5016 - mse: 0.4175\n",
      "Epoch 23/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4166 - mae: 0.4911 - mse: 0.4166\n",
      "Epoch 24/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4155 - mae: 0.4918 - mse: 0.4155\n",
      "Epoch 25/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4147 - mae: 0.4931 - mse: 0.4147\n",
      "Epoch 26/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4144 - mae: 0.4922 - mse: 0.4144\n",
      "Epoch 27/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4141 - mae: 0.4902 - mse: 0.4141\n",
      "Epoch 28/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4141 - mae: 0.4900 - mse: 0.4141\n",
      "Epoch 29/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4923 - mse: 0.4139\n",
      "Epoch 30/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4917 - mse: 0.4138\n",
      "Epoch 31/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4893 - mse: 0.4137\n",
      "Epoch 32/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4854 - mse: 0.4139\n",
      "Epoch 33/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4882 - mse: 0.4139\n",
      "Epoch 34/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4898 - mse: 0.4139\n",
      "Epoch 35/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4816 - mse: 0.4137\n",
      "Epoch 36/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4933 - mse: 0.4137\n",
      "Epoch 37/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4883 - mse: 0.4137\n",
      "Epoch 38/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4132 - mae: 0.4801 - mse: 0.4132\n",
      "Epoch 39/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4855 - mse: 0.4137\n",
      "Epoch 40/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4135 - mae: 0.4975 - mse: 0.4135\n",
      "Epoch 41/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4905 - mse: 0.4137\n",
      "Epoch 42/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4851 - mse: 0.4138\n",
      "Epoch 43/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4132 - mae: 0.4889 - mse: 0.4132\n",
      "Epoch 44/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4928 - mse: 0.4137\n",
      "Epoch 45/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4864 - mse: 0.4136\n",
      "Epoch 46/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4140 - mae: 0.4923 - mse: 0.4140\n",
      "Epoch 47/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4837 - mse: 0.4138\n",
      "Epoch 48/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4909 - mse: 0.4138\n",
      "Epoch 49/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4855 - mse: 0.4137\n",
      "Epoch 50/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4883 - mse: 0.4138\n",
      "Epoch 51/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4890 - mse: 0.4139\n",
      "Epoch 52/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4979 - mse: 0.4138\n",
      "Epoch 53/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4136 - mae: 0.4861 - mse: 0.4136\n",
      "Epoch 54/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4132 - mae: 0.4830 - mse: 0.4132\n",
      "Epoch 55/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4130 - mae: 0.4840 - mse: 0.4130\n",
      "Epoch 56/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4135 - mae: 0.4892 - mse: 0.4135\n",
      "Epoch 57/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4133 - mae: 0.4945 - mse: 0.4133\n",
      "Epoch 58/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4131 - mae: 0.4946 - mse: 0.4131\n",
      "Epoch 59/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4135 - mae: 0.4834 - mse: 0.4135\n",
      "Epoch 60/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4922 - mse: 0.4139\n",
      "Epoch 61/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4860 - mse: 0.4138\n",
      "Epoch 62/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4134 - mae: 0.4980 - mse: 0.4134\n",
      "Epoch 63/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4926 - mse: 0.4137\n",
      "Epoch 64/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4140 - mae: 0.4892 - mse: 0.4140\n",
      "Epoch 65/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4129 - mae: 0.4811 - mse: 0.4129\n",
      "Epoch 66/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4914 - mse: 0.4138\n",
      "Epoch 67/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4957 - mse: 0.4137\n",
      "Epoch 68/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4915 - mse: 0.4138\n",
      "Epoch 69/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4129 - mae: 0.4910 - mse: 0.4129\n",
      "Epoch 70/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4919 - mse: 0.4138\n",
      "Epoch 71/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4926 - mse: 0.4137\n",
      "Epoch 72/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4884 - mse: 0.4138\n",
      "Epoch 73/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4141 - mae: 0.4875 - mse: 0.4141\n",
      "Epoch 74/200\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4136 - mae: 0.4875 - mse: 0.4136\n",
      "Epoch 75/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4810 - mse: 0.4136\n",
      "Epoch 76/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4874 - mse: 0.4140\n",
      "Epoch 77/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4874 - mse: 0.4140\n",
      "Epoch 78/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4810 - mse: 0.4135\n",
      "Epoch 79/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4840 - mse: 0.4137\n",
      "Epoch 80/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4855 - mse: 0.4138\n",
      "Epoch 81/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4132 - mae: 0.4876 - mse: 0.4132\n",
      "Epoch 82/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4899 - mse: 0.4138\n",
      "Epoch 83/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4136 - mae: 0.4850 - mse: 0.4136\n",
      "Epoch 84/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4958 - mse: 0.4137\n",
      "Epoch 85/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4924 - mse: 0.4138\n",
      "Epoch 86/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4956 - mse: 0.4135\n",
      "Epoch 87/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4132 - mae: 0.4832 - mse: 0.4132\n",
      "Epoch 88/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4920 - mse: 0.4139\n",
      "Epoch 89/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4136 - mae: 0.4900 - mse: 0.4136\n",
      "Epoch 90/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4911 - mse: 0.4139\n",
      "Epoch 91/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4846 - mse: 0.4137\n",
      "Epoch 92/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4901 - mse: 0.4140\n",
      "Epoch 93/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4845 - mse: 0.4137\n",
      "Epoch 94/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4134 - mae: 0.4819 - mse: 0.4134\n",
      "Epoch 95/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4921 - mse: 0.4139\n",
      "Epoch 96/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4903 - mse: 0.4139\n",
      "Epoch 97/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4885 - mse: 0.4135\n",
      "Epoch 98/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4818 - mse: 0.4137\n",
      "Epoch 99/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4875 - mse: 0.4138\n",
      "Epoch 100/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4857 - mse: 0.4137\n",
      "Epoch 101/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4911 - mse: 0.4139\n",
      "Epoch 102/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4904 - mse: 0.4139\n",
      "Epoch 103/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4911 - mse: 0.4138\n",
      "Epoch 104/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4912 - mse: 0.4140\n",
      "Epoch 105/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4946 - mse: 0.4139\n",
      "Epoch 106/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4861 - mse: 0.4139\n",
      "Epoch 107/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4843 - mse: 0.4135\n",
      "Epoch 108/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4863 - mse: 0.4136\n",
      "Epoch 109/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4938 - mse: 0.4139\n",
      "Epoch 110/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4134 - mae: 0.4853 - mse: 0.4134\n",
      "Epoch 111/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4918 - mse: 0.4138\n",
      "Epoch 112/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4904 - mse: 0.4139\n",
      "Epoch 113/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4854 - mse: 0.4135\n",
      "Epoch 114/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4937 - mse: 0.4139\n",
      "Epoch 115/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4901 - mse: 0.4140\n",
      "Epoch 116/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4897 - mse: 0.4138\n",
      "Epoch 117/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4881 - mse: 0.4139\n",
      "Epoch 118/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4906 - mse: 0.4140\n",
      "Epoch 119/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4877 - mse: 0.4139\n",
      "Epoch 120/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4859 - mse: 0.4136\n",
      "Epoch 121/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4972 - mse: 0.4136\n",
      "Epoch 122/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4900 - mse: 0.4139\n",
      "Epoch 123/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4132 - mae: 0.4932 - mse: 0.4132\n",
      "Epoch 124/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4805 - mse: 0.4135\n",
      "Epoch 125/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4132 - mae: 0.4915 - mse: 0.4132\n",
      "Epoch 126/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4916 - mse: 0.4139\n",
      "Epoch 127/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4923 - mse: 0.4138\n",
      "Epoch 128/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4930 - mse: 0.4138\n",
      "Epoch 129/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4863 - mse: 0.4139\n",
      "Epoch 130/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4908 - mse: 0.4140\n",
      "Epoch 131/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4875 - mse: 0.4138\n",
      "Epoch 132/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4131 - mae: 0.4763 - mse: 0.4131\n",
      "Epoch 133/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4887 - mse: 0.4139\n",
      "Epoch 134/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4131 - mae: 0.4830 - mse: 0.4131\n",
      "Epoch 135/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4861 - mse: 0.4137\n",
      "Epoch 136/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4861 - mse: 0.4135\n",
      "Epoch 137/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4842 - mse: 0.4137\n",
      "Epoch 138/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4141 - mae: 0.4924 - mse: 0.4141\n",
      "Epoch 139/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4948 - mse: 0.4138\n",
      "Epoch 140/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4842 - mse: 0.4136\n",
      "Epoch 141/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4956 - mse: 0.4137\n",
      "Epoch 142/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4872 - mse: 0.4137\n",
      "Epoch 143/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4895 - mse: 0.4139\n",
      "Epoch 144/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4865 - mse: 0.4136\n",
      "Epoch 145/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4940 - mse: 0.4137\n",
      "Epoch 146/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4884 - mse: 0.4139\n",
      "Epoch 147/200\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4130 - mae: 0.4828 - mse: 0.4130\n",
      "Epoch 148/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4875 - mse: 0.4140\n",
      "Epoch 149/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4133 - mae: 0.4924 - mse: 0.4133\n",
      "Epoch 150/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4875 - mse: 0.4137\n",
      "Epoch 151/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4866 - mse: 0.4138\n",
      "Epoch 152/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4133 - mae: 0.4924 - mse: 0.4133\n",
      "Epoch 153/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4889 - mse: 0.4140\n",
      "Epoch 154/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4895 - mse: 0.4140\n",
      "Epoch 155/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4835 - mse: 0.4138\n",
      "Epoch 156/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4862 - mse: 0.4139\n",
      "Epoch 157/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4868 - mse: 0.4138\n",
      "Epoch 158/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4910 - mse: 0.4138\n",
      "Epoch 159/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4858 - mse: 0.4138\n",
      "Epoch 160/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4881 - mse: 0.4139\n",
      "Epoch 161/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4895 - mse: 0.4140\n",
      "Epoch 162/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4131 - mae: 0.4949 - mse: 0.4131\n",
      "Epoch 163/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4830 - mse: 0.4138\n",
      "Epoch 164/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4943 - mse: 0.4135\n",
      "Epoch 165/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4824 - mse: 0.4138\n",
      "Epoch 166/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4889 - mse: 0.4140\n",
      "Epoch 167/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4134 - mae: 0.4799 - mse: 0.4134\n",
      "Epoch 168/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4864 - mse: 0.4137\n",
      "Epoch 169/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4134 - mae: 0.4937 - mse: 0.4134\n",
      "Epoch 170/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4894 - mse: 0.4140\n",
      "Epoch 171/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4887 - mse: 0.4137\n",
      "Epoch 172/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4906 - mse: 0.4139\n",
      "Epoch 173/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4860 - mse: 0.4135\n",
      "Epoch 174/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4947 - mse: 0.4135\n",
      "Epoch 175/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4869 - mse: 0.4139\n",
      "Epoch 176/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4864 - mse: 0.4135\n",
      "Epoch 177/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4841 - mse: 0.4138\n",
      "Epoch 178/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4133 - mae: 0.4857 - mse: 0.4133\n",
      "Epoch 179/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4915 - mse: 0.4140\n",
      "Epoch 180/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4901 - mse: 0.4139\n",
      "Epoch 181/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4932 - mse: 0.4137\n",
      "Epoch 182/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4887 - mse: 0.4138\n",
      "Epoch 183/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4905 - mse: 0.4137\n",
      "Epoch 184/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4861 - mse: 0.4138\n",
      "Epoch 185/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4136 - mae: 0.4954 - mse: 0.4136\n",
      "Epoch 186/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4933 - mse: 0.4139\n",
      "Epoch 187/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4905 - mse: 0.4139\n",
      "Epoch 188/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4797 - mse: 0.4135\n",
      "Epoch 189/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4140 - mae: 0.4899 - mse: 0.4140\n",
      "Epoch 190/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4138 - mae: 0.4942 - mse: 0.4138\n",
      "Epoch 191/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4139 - mae: 0.4856 - mse: 0.4139\n",
      "Epoch 192/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4137 - mae: 0.4908 - mse: 0.4137\n",
      "Epoch 193/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4914 - mse: 0.4139\n",
      "Epoch 194/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4909 - mse: 0.4139\n",
      "Epoch 195/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4894 - mse: 0.4139\n",
      "Epoch 196/200\n",
      "10/10 [==============================] - 0s 1ms/sample - loss: 0.4135 - mae: 0.4965 - mse: 0.4135\n",
      "Epoch 197/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4138 - mae: 0.4930 - mse: 0.4138\n",
      "Epoch 198/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4136 - mae: 0.4804 - mse: 0.4136\n",
      "Epoch 199/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4137 - mae: 0.4902 - mse: 0.4137\n",
      "Epoch 200/200\n",
      "10/10 [==============================] - 0s 2ms/sample - loss: 0.4139 - mae: 0.4925 - mse: 0.4139\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f4013d4e950>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train_norm, y_train, \n",
    "          epochs=num_epochs, batch_size=batch_size,\n",
    "          verbose=1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.7058775 4.971019\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 936x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(model.w.numpy(), model.b.numpy())\n",
    "\n",
    "\n",
    "X_test = np.linspace(0, 9, num=100).reshape(-1, 1)\n",
    "X_test_norm = (X_test - np.mean(X_train)) / np.std(X_train)\n",
    "\n",
    "y_pred = model(tf.cast(X_test_norm, dtype=tf.float32))\n",
    "\n",
    "\n",
    "fig = plt.figure(figsize=(13, 5))\n",
    "ax = fig.add_subplot(1, 2, 1)\n",
    "plt.plot(X_train_norm, y_train, 'o', markersize=10)\n",
    "plt.plot(X_test_norm, y_pred, '--', lw=3)\n",
    "plt.legend(['Training Samples', 'Linear Regression'], fontsize=15)\n",
    "\n",
    "ax = fig.add_subplot(1, 2, 2)\n",
    "plt.plot(Ws, lw=3)\n",
    "plt.plot(bs, lw=3)\n",
    "plt.legend(['W', 'bias'], fontsize=15)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building a multilayer perceptron for classifying flowers in the Iris dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tfds.core.DatasetInfo(\n",
      "    name='iris',\n",
      "    version=1.0.0,\n",
      "    description='This is perhaps the best known database to be found in the pattern recognition\n",
      "literature. Fisher's paper is a classic in the field and is referenced\n",
      "frequently to this day. (See Duda & Hart, for example.) The data set contains\n",
      "3 classes of 50 instances each, where each class refers to a type of iris\n",
      "plant. One class is linearly separable from the other 2; the latter are NOT\n",
      "linearly separable from each other.\n",
      "',\n",
      "    urls=['https://archive.ics.uci.edu/ml/datasets/iris'],\n",
      "    features=FeaturesDict({\n",
      "        'features': Tensor(shape=(4,), dtype=tf.float32),\n",
      "        'label': ClassLabel(shape=(), dtype=tf.int64, num_classes=3),\n",
      "    }),\n",
      "    total_num_examples=150,\n",
      "    splits={\n",
      "        'train': 150,\n",
      "    },\n",
      "    supervised_keys=('features', 'label'),\n",
      "    citation=\"\"\"@misc{Dua:2019 ,\n",
      "    author = \"Dua, Dheeru and Graff, Casey\",\n",
      "    year = \"2017\",\n",
      "    title = \"{UCI} Machine Learning Repository\",\n",
      "    url = \"http://archive.ics.uci.edu/ml\",\n",
      "    institution = \"University of California, Irvine, School of Information and Computer Sciences\"\n",
      "    }\"\"\",\n",
      "    redistribution_info=,\n",
      ")\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import tensorflow_datasets as tfds\n",
    "\n",
    "\n",
    "\n",
    "iris, iris_info = tfds.load('iris', with_info=True)\n",
    "\n",
    "print(iris_info)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'features': <tf.Tensor: id=117438, shape=(4,), dtype=float32, numpy=array([5.5, 3.5, 1.3, 0.2], dtype=float32)>, 'label': <tf.Tensor: id=117439, shape=(), dtype=int64, numpy=0>}\n"
     ]
    }
   ],
   "source": [
    "tf.random.set_seed(1)\n",
    "\n",
    "ds_orig = iris['train']\n",
    "ds_orig = ds_orig.shuffle(150, reshuffle_each_iteration=False)\n",
    "\n",
    "print(next(iter(ds_orig)))\n",
    "\n",
    "ds_train_orig = ds_orig.take(100)\n",
    "ds_test = ds_orig.skip(100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100\n",
      "50\n"
     ]
    }
   ],
   "source": [
    "## checking the number of examples:\n",
    "\n",
    "n = 0\n",
    "for example in ds_train_orig:\n",
    "    n += 1\n",
    "print(n)\n",
    "\n",
    "\n",
    "n = 0\n",
    "for example in ds_test:\n",
    "    n += 1\n",
    "print(n)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(<tf.Tensor: id=117773, shape=(4,), dtype=float32, numpy=array([5.5, 3.5, 1.3, 0.2], dtype=float32)>,\n",
       " <tf.Tensor: id=117774, shape=(), dtype=int64, numpy=0>)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ds_train_orig = ds_train_orig.map(\n",
    "    lambda x: (x['features'], x['label']))\n",
    "\n",
    "ds_test = ds_test.map(\n",
    "    lambda x: (x['features'], x['label']))\n",
    "\n",
    "next(iter(ds_train_orig))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "fc1 (Dense)                  (None, 16)                80        \n",
      "_________________________________________________________________\n",
      "fc2 (Dense)                  (None, 3)                 51        \n",
      "=================================================================\n",
      "Total params: 131\n",
      "Trainable params: 131\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model = tf.keras.Sequential([\n",
    "    tf.keras.layers.Dense(16, activation='sigmoid', \n",
    "                          name='fc1', input_shape=(4,)),\n",
    "    tf.keras.layers.Dense(3, name='fc2', activation='softmax')])\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer='adam',\n",
    "              loss='sparse_categorical_crossentropy',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_epochs = 100\n",
    "training_size = 100\n",
    "batch_size = 2\n",
    "steps_per_epoch = np.ceil(training_size / batch_size)\n",
    "\n",
    "ds_train = ds_train_orig.shuffle(buffer_size=training_size)\n",
    "ds_train = ds_train.repeat()\n",
    "ds_train = ds_train.batch(batch_size=batch_size)\n",
    "ds_train = ds_train.prefetch(buffer_size=1000)\n",
    "\n",
    "\n",
    "history = model.fit(ds_train, epochs=num_epochs,\n",
    "                    steps_per_epoch=steps_per_epoch, \n",
    "                    verbose=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "hist = history.history\n",
    "\n",
    "fig = plt.figure(figsize=(12, 5))\n",
    "ax = fig.add_subplot(1, 2, 1)\n",
    "ax.plot(hist['loss'], lw=3)\n",
    "ax.set_title('Training loss', size=15)\n",
    "ax.set_xlabel('Epoch', size=15)\n",
    "ax.tick_params(axis='both', which='major', labelsize=15)\n",
    "\n",
    "ax = fig.add_subplot(1, 2, 2)\n",
    "ax.plot(hist['accuracy'], lw=3)\n",
    "ax.set_title('Training accuracy', size=15)\n",
    "ax.set_xlabel('Epoch', size=15)\n",
    "ax.tick_params(axis='both', which='major', labelsize=15)\n",
    "plt.tight_layout()\n",
    "#plt.savefig('ch13-cls-learning-curve.pdf')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluating the trained model on the test dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss: 0.1491   Test Acc.: 1.0000\n"
     ]
    }
   ],
   "source": [
    "results = model.evaluate(ds_test.batch(50), verbose=0)\n",
    "print('Test loss: {:.4f}   Test Acc.: {:.4f}'.format(*results))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving and reloading the trained model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('iris-classifier.h5', \n",
    "           overwrite=True,\n",
    "           include_optimizer=True,\n",
    "           save_format='h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "fc1 (Dense)                  (None, 16)                80        \n",
      "_________________________________________________________________\n",
      "fc2 (Dense)                  (None, 3)                 51        \n",
      "=================================================================\n",
      "Total params: 131\n",
      "Trainable params: 131\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_new = tf.keras.models.load_model('iris-classifier.h5')\n",
    "\n",
    "model_new.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test loss: 0.1491   Test Acc.: 1.0000\n"
     ]
    }
   ],
   "source": [
    "results = model_new.evaluate(ds_test.batch(50), verbose=0)\n",
    "print('Test loss: {:.4f}   Test Acc.: {:.4f}'.format(*results))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Set:  100 Test Set:  50\n"
     ]
    }
   ],
   "source": [
    "labels_train = []\n",
    "for i,item in enumerate(ds_train_orig):\n",
    "    labels_train.append(item[1].numpy())\n",
    "    \n",
    "labels_test = []\n",
    "for i,item in enumerate(ds_test):\n",
    "    labels_test.append(item[1].numpy())\n",
    "print('Training Set: ',len(labels_train), 'Test Set: ', len(labels_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'{\"class_name\": \"Sequential\", \"config\": {\"name\": \"sequential\", \"layers\": [{\"class_name\": \"Dense\", \"config\": {\"name\": \"fc1\", \"trainable\": true, \"batch_input_shape\": [null, 4], \"dtype\": \"float32\", \"units\": 16, \"activation\": \"sigmoid\", \"use_bias\": true, \"kernel_initializer\": {\"class_name\": \"GlorotUniform\", \"config\": {\"seed\": null}}, \"bias_initializer\": {\"class_name\": \"Zeros\", \"config\": {}}, \"kernel_regularizer\": null, \"bias_regularizer\": null, \"activity_regularizer\": null, \"kernel_constraint\": null, \"bias_constraint\": null}}, {\"class_name\": \"Dense\", \"config\": {\"name\": \"fc2\", \"trainable\": true, \"dtype\": \"float32\", \"units\": 3, \"activation\": \"softmax\", \"use_bias\": true, \"kernel_initializer\": {\"class_name\": \"GlorotUniform\", \"config\": {\"seed\": null}}, \"bias_initializer\": {\"class_name\": \"Zeros\", \"config\": {}}, \"kernel_regularizer\": null, \"bias_regularizer\": null, \"activity_regularizer\": null, \"kernel_constraint\": null, \"bias_constraint\": null}}]}, \"keras_version\": \"2.2.4-tf\", \"backend\": \"tensorflow\"}'"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.to_json()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Choosing activation functions for multilayer neural networks\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Logistic function recap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "P(y=1|x) = 0.888\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "X = np.array([1, 1.4, 2.5]) ## first value must be 1\n",
    "w = np.array([0.4, 0.3, 0.5])\n",
    "\n",
    "def net_input(X, w):\n",
    "    return np.dot(X, w)\n",
    "\n",
    "def logistic(z):\n",
    "    return 1.0 / (1.0 + np.exp(-z))\n",
    "\n",
    "def logistic_activation(X, w):\n",
    "    z = net_input(X, w)\n",
    "    return logistic(z)\n",
    "\n",
    "print('P(y=1|x) = %.3f' % logistic_activation(X, w)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Net Input: \n",
      " [1.78 0.76 1.65]\n",
      "Output Units:\n",
      " [0.85569687 0.68135373 0.83889105]\n"
     ]
    }
   ],
   "source": [
    "# W : array with shape = (n_output_units, n_hidden_units+1)\n",
    "# note that the first column are the bias units\n",
    "\n",
    "W = np.array([[1.1, 1.2, 0.8, 0.4],\n",
    "              [0.2, 0.4, 1.0, 0.2],\n",
    "              [0.6, 1.5, 1.2, 0.7]])\n",
    "\n",
    "# A : data array with shape = (n_hidden_units + 1, n_samples)\n",
    "# note that the first column of this array must be 1\n",
    "\n",
    "A = np.array([[1, 0.1, 0.4, 0.6]])\n",
    "Z = np.dot(W, A[0])\n",
    "y_probas = logistic(Z)\n",
    "print('Net Input: \\n', Z)\n",
    "\n",
    "print('Output Units:\\n', y_probas) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted class label: 0\n"
     ]
    }
   ],
   "source": [
    "y_class = np.argmax(Z, axis=0)\n",
    "print('Predicted class label: %d' % y_class) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Estimating class probabilities in multiclass classification via the softmax function"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Probabilities:\n",
      " [0.44668973 0.16107406 0.39223621]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "1.0"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def softmax(z):\n",
    "    return np.exp(z) / np.sum(np.exp(z))\n",
    "\n",
    "y_probas = softmax(Z)\n",
    "print('Probabilities:\\n', y_probas)\n",
    "\n",
    "np.sum(y_probas)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=134666, shape=(1, 3), dtype=float64, numpy=array([[0.44668973, 0.16107406, 0.39223621]])>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "Z_tensor = tf.expand_dims(Z, axis=0)\n",
    "tf.keras.activations.softmax(Z_tensor)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Broadening the output spectrum using a hyperbolic tangent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEYCAYAAAAJeGK1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nO3de1xUdfrA8c/DHQRF8Z6alyw1U0vL7LKd0kq72Ja1YVItbUu7UWm/ddvcSttsd2mzNrpYZltaVNqWXdyKkjLKyAILFSVSEQUVFS8oIDDA9/fHGUcQEJDLDPC8X695zXnOZeZhHHn4nvM9368YY1BKKaU8jZe7E1BKKaVqogVKKaWUR9ICpZRSyiNpgVJKKeWRtEAppZTySFqglFJKeSSPKlAi8qqI7BGRtFq2WyKSLyKpzsfsls5RKaVUy/BxdwLHWQQ8D7x+gn2+McZc0zLpKKWUchePakEZY74G9rs7D6WUUu7naS2o+hgnImuBncBMY8yGmnYSkSggCqBDhw6jhwwZ0oIpKqWUqq81a9bkGWO6Hb++tRWoH4FTjTEFInIV8AEwuKYdjTEvAy8DjBkzxqSkpLRclkoppepNRLbVtN6jTvHVxRhzyBhT4Fz+BPAVka5uTksppVQzaFUFSkR6iog4l8/Dzn+fe7NSSinVHDzqFJ+IvA1YQFcRyQHmAL4AxpiXgBuBP4pIGXAECDc6HLtSSrVJHlWgjDFT69j+PHY3dKWUUm1cqzrFp5RSqv3QAqWUUsojaYFSSinlkbRAKaWU8khaoJRSSnkkLVBKKaU8khYopZRSHkkLlFJKKY+kBUoppZRH0gKllFLKI2mBUkop5ZG0QCmllPJIWqCUUkp5JC1QSimlPJIWKKWUUh5JC5RSSimPpAVKKaWUR9ICpZRSyiNpgVJKKeWRtEAppZTySFqglFJKeSQtUEoppTySFiillFIeSQuUUkopj6QFSimllEfSAqWUUsojaYFSSinlkbRAKaWU8khaoJRSSnkkjypQIvKqiOwRkbRatouIPCsim0VknYic09I5KqWUahk+7k7gOIuA54HXa9k+CRjsfIwFXnQ+K6WURzPGUFZhKD/6MAZjAAMGQ4Wx9/Hz8SIkwLfKsflHHBSWlGGcr2OMvb7CuexaD3QK9KVrsH+V43cePMKhYkelXKhxGaB7R/9qx2/NK6SwpKzGn2tQt2AC/bwb8EnUn0e1oIwxXwP7T7DLdcDrxrYaCBWRXnW9bkZGBosWLQLA4XBgWRZxcXEAFBUVYVkWS5cuBSA/Px/Lsli2bBkAeXl5WJbF8uXLAcjNzcWyLOLj4wHIzs7GsiwSEhIAyMzMxLIsEhMTXe9tWRZJSUkApKWlYVkWycnJAKSmpmJZFqmpqQAkJydjWRZpaXYjMikpCcuyyMjIACAxMRHLssjMzAQgISEBy7LIzs4GID4+HsuyyM3NBWD58uVYlkVeXh4Ay5Ytw7Is8vPzAVi6dCmWZVFUVARAXFwclmXhcNhf5kWLFmFZluuzXLhwIRMmTHDF8+fPZ9KkSa44NjaWyZMnu+J58+YxZcoUVxwTE0N4eLgrnjt3LhEREa549uzZREZGuuJZs2YRFRXlimfOnEl0dLQrnjFjBjNmzHDF0dHRzJw50xVHRUUxa9YsVxwZGcns2bNdcUREBHPnznXF4eHhxMTEuOIpU6Ywb948Vzx58mRiY2Nd8aRJk5g/f74rnjBhAgsXLnTFlmXpd8+Dvns3h4dzoLCUbfsKmfHoPK69436+/Hk3H6buIPyh5xgf/Q/e+n47UPW7l7Qlj3EPLGbMzEX8ZsF3XPfCt4x4YAnDH3iHS55cyYUxXzLkgXcZ8pdl3PVGClD1u7cgcQsDZn3C4Ic+Zcgj8Zw55zNGPPo5I//2OSMf+5xRj63gnLkrGP14An993/78K3/3nvtiExfEfMmFMV9y0RMrufhf9uOSJ7/CmvcVl877isueSmT8U4ksWJkBJQVMnnAR77zyDOzP5NVlH/On2Nf5a+x/+NuzL/HEc8/xzHNP8+JzT/DaC3N584U5vPPCQyx74UG2LJtL4adzWPzbM9j0UgR89hAbXrmLH1/8HetevJ30FyPY9NJUsl76DTkvTaEs7ib2f/hQo757tfG0FlRdTgGyK8U5znW7jt9RRKKAKAB/f//jNyul2hBHeQX5ZT4UBPXii/TdjB/ao8r2VZvyWJw/hCOnnsnZc1c41w6F7kO5Y1GKMx4IIfDqt1u5ZWy/KsfvOVTCLq+u4AV5W51/Q3uFAFCwr8gZBwJwoMjB8by9pJbMDUGU0JFCOkoRnSjkzMNbYW0GV3bJ4YzSJFj5DyZkb2eAzx6CpJggSgiimA6VloOkhEBK8MeBz5oKWAMfXQTkrIdn5/AwQH1/DWbaj9v7A7nLIReugdqrxXYoNBX1fPGGEXN8+87NRKQ/8D9jzPAatv0PiDHGrHLGXwB/McakHL9vZWPGjDEpKSfcRSnVCuQVlPBL7mE27y1gy54CtuwtZMveAnIPFbtOVXUK9GXtnCuqHJectZ+bXvquXu/Rq1MA380aX2Xdp+t38cc3f6zjSENHCvnVKcLzk/tC4V77UbSPjZu2kLkti05SSKizGIVQSAhF+Ep5fX98zzX4Spj2zkkfLiJrjDFjjl/f2lpQO4C+leI+znVKqTZsy94Cbn3le3bmF9e5b/4RBwUlZQT7H/v11qWDn2s5JMCHToG+BPv72I8AHzr4+xDsZy+HBftVe81zTgkk7obuBJfspkPJboKKdxNwZDf+RbvwLczFu3A33kf2IRUO2Ae8VvX4YcCw5rlMUwMB30Dw9gOfAPDxdz4fjZ3rvP2rbvPyBW9f8PJxPvuCt0+l9cfHlfYL6VF3WiehtRWoj4B7RGQJdueIfGNMtdN7SqnWp6LCkJpzkG835XHnxQOrXHjv3SmQ3YdLaj1WBLqH+NM7NJDeoYEUO8qrFKhTuwSxetZ4Onfwxd+nhkphDBTshv1b4cBW+HKp/bx/KxzIokdRHs3zKxjwCYTAUAjoBAFHnzuCfwj4dQDfDvazXxD4BYNvkDMOdq5z7uMbaD+8fOwPpA3wqAIlIm8DFtBVRHKAOYAvgDHmJeAT4CpgM1AERNb8Skqp1qCsvILvt+7n07RdfLZhN3udRWhUv1AuHtzNtV+gnzeDuwezNa+QIb06Mrh7MIO6BTOoWwdO6x5Mn85B+PnU3ufLx9uLnp0CoNwBe7fA3p9hbwbsTbef922BsiON/4H8gqFDV+jQzX4EhR1b7tAVArscV4w62q0YVSOPKlDGmKl1bDdA9In2UUp5vq15hSxNzubdNTnkFVRvGX27eV+VAgWwKPI8ugb74eNdz87HRfth19pjj70/Q94mqKjeiaFO4g0hvaBjb+fjlKrLIT3tIuQX1PDXVrXyqAKllGq7jDGszNjDwq+38l3mvhr3Cevgx6VDunP+wC7VtvXsFFD7i5cchpwU2Pkj7Ey1C9LBbQ1LMCAUugyAzv2h8wDnsjPu2Bu8WuwiknLSAqWUahHPfbmZp1f8Um19txB/rhrek4nDe3HegC4n6JJdSf4OyF4N252P3WlQ367OHU+BbmdAt6HO5yHQdTAEVS+Kyr20QCmlWsT1Z59C7BebKK8weAlcNqQ74ef2wzqjW92n7Qr2QGYiZH4FW7+G/O11v6G3H3QfBr1GQu9R0OMsuyAFdGySn0c1Py1QSqkml7m3gL5dgvCtVHj6dgniN2P6EuTnze8uGkDv0MDaX8BxBLK+hcyVdlHaXePwnJUI9DgT+pwLvc+2C1K3oXb3adVqaYFSSjWZkrJyXvxqCy+s3Mz9l5/O3dZpVbb/84azaj+4YA/88hlkfApbvjxxrzrfIDhlNPQbB/3G2oUpoFMT/RTKU2iBUko1ifRdh7jv7Z/YtKcAgNiETUwa3osBXTvUflDeZkj/yC5KOcnYw57WwMsX+p4HAy370fts+yZR1aZpgVJKNYoxhrjvtzP3fxspLTvWUWFor45U1DSU2sHtsOF9WP8u5K6r/YW7ng6nXW4XpFMvAP/gJs9deTYtUEqpk1ZUWsaf/7uOj9cfG9Al0NebByaewW3j+h/rkVewFzYsg7T3IPv7ml9MvOxTdmdcBWdMgrBBLfATKE+mBUopdVJy84v53eJkNuw85Fo3pGcIz99yDqd1D4byMvjlC/jxdfglHipqmE/I289uJQ2bDIOv0K7eqgotUEqpBkvbkc/vFiez+9CxUSCmje3HI9cMI+BQFiTMg9S3oSC3+sHibZ+2O+tGGHK1dm5QtdICpZRqkJKycqJeT3EVJx8vYe61Q5gauhHe+qt9n1JN+pwHI8Nh2HX2uHRK1UELlFKqQfx9vImdeja3/ecHunkf5vWzf6H/6gcgP7v6zh26w6ipMCoCup3e8smqVk0LlFKqwc4N2MlXp/+XblnL8frpuMFexQtOnwhn3wqDL9fu4OqkaYFSStXJGIMAZK2Cb5+BzQnV50cKCoNzbocxd0Bo3+ovolQDaYFSSp1Qxq58lr31Ev8X9Cn+u3+qvkOvUTD2LjjzBvA9wYjjSjWQFiilVM3KHeSvfp2AFfOYxU44XHmj2F3Dx91jDzPURmZwVZ5FC5RSqqpyB6x9m4rEJ+mUv53KncArvP3xGjUVLrhPb6RVzU4LlFLKVl4G65bA10/CgSwqT4BxyASRf+Zt9J30fxBS7eqTUs1CC5RS7V15Gax/BxL/BQe2Vtm03wSzsOwa+k28l6kXD3dTgqq90gKlVHtlDGR8Aitmw77NVTYdMMEsLLuaxeVXcP35Z2hxUm6hBUqp9mjXWvjsIcj6psrqioBQFjiu4vnC8RQSyLn9OzPn2jPdlKRq77RAKdWeHNoFX86F1LeoMveSfycqxt3D3VvOJX5TEQCdg3x5durZVWbFVaolaYFSqj0oLYSk5+2bbB1Fx9aLN5x7J1zyF/67sZD4Tetdm57+zSh6dTrBtOxKNTMtUEq1ZRUVsG4pfPEYHN5Zddvpk+Dyx1xj5F09ohPrcvJ58/vtRP1qIJcO6e6GhJU6RguUUm1V1ir47K/29abKegyHKx6HQZdWWR3s78Pfrz+La0f25ux+oS2YqFI10wKlVFuzb4vdM+/n/1Vd36E7jH8ERk0DL+9aDz9/YFgzJ6hU/WiBUqqtOHIAEp+EH16GCsex9T4BcMG9cOF08A+pckhZeQU+2glCeSgtUEq1duUOSP4PJMbYRaqyETfD+NnQqU+1w4wx/CFuDZ2D/Hhw0hDCgv1bKGGl6kcLlFKtlTGQ8SmseKTajbb0GwdX/h1OGV3r4Ss27iYhfQ8ACem7+WrmpXQK0rmblOfwqLa9iEwUkQwR2SwiD9aw/bcisldEUp2PO92Rp1Jut2sdLL4WlkytWpw694ffvA6Rn56wOBWVlvG35Rtd8aSzemlxUh7HY1pQIuINvABcDuQAySLykTFm43G7LjXG3NPiCSrlCQ7tgpWPw09vcvyNtlzyZzgvCnzqPlX33Jeb2XHwCABdOvjxwJVnNFPCSp08T2pBnQdsNsZkGmNKgSXAdU3xwhkZGSxatAgAh8OBZVnExcUBUFRUhGVZLF26FID8/Hwsy2LZsmUA5OXlYVkWy5cvByA3NxfLsoiPjwcgOzsby7JISEgAIDMzE8uySExMdL23ZVkkJSUBkJaWhmVZJCcnA5CamoplWaSmpgKQnJyMZVmkpaUBkJSUhGVZZGRkAJCYmIhlWWRmZgKQkJCAZVlkZ2cDEB8fj2VZ5ObmArB8+XIsyyIvLw+AZcuWYVkW+fn5ACxduhTLsigqsm/ejIuLw7IsHA77IvuiRYuwLMv1WS5cuJAJEya44vnz5zNp0iRXHBsby+TJk13xvHnzmDJliiuOiYkhPDzcFc+dO5eIiAhXPHv2bCIjI13xrFmziIqKcsUzZ84kOjraFc+YMYMZM2a44ujoaGbOnOmKo6KimDVrliuOjIxk9uzZrjgiIoK5c+e64vDwcGJiYlzxlClTmDdvniuePHkysbGxrnjSpEnMnz/fFU+YMIGFCxe6Ysuymua7V1rE4Y8f4ciTw+CnOI4Wp3ID2b0mwX0/kdnzaqwJV9b53fto5Xe88k2mK8dpZwYRGuSn3z397rnt915tPKlAnQJkV4pznOuON0VE1onIuyJS67zSIhIlIikiknL0C69UayMY+uxbBc+NJiT5WQK9K1zbjvT9FZHJQ8kYdCd0qH/X8LfSCnGU2wXO//AOLh0Q1OR5K9UUxBhT914tQERuBCYaY+50xrcCYyufzhORMKDAGFMiIncBNxtjLqvrtceMGWNSUlKaK3WlmkfWt84bbY/7K7P7mXDl4zCozq9+NWu27WfKi9+54vfvvoCz+3VubKZKNYqIrDHGjDl+vcdcgwJ2AJVbRH2c61yMMfsqha8A/2qBvJRqWfu2QMIcSF9edX2H7nDZw3B2xAlvtK2NMYbHP053xfaIEVqclOdqcIESkQ5AsTGmvIlzSQYGi8gA7MIUDtxy3Hv3MsbscoaTgXSUaiuOHICv58H3C6rfaDvuHrhoRrUbbRvik/W5/LT9IAB+3l7aMUJ5vDoLlIh4YReLacC5QAngLyJ5wMfAAmPM5hO8RL0YY8pE5B7gM8AbeNUYs0FEHgNSjDEfAfeJyGSgDNgP/Lax76uU25WVQsp/IPGJ6jfanvUb+0bb0Fovt9bb5xtzXcu/vbA/fbvotSfl2eq8BiUiiUAC8CGQZoypcK7vAlyK3cp53xgT18y5njS9BqU8kjGQ/hGsmFNtqnX6ng9X/gP61H4vU8PfzhCflstLX2fyeuR5et+T8hi1XYOqT4HyNcacsBtcffZxJy1QyuNkJ8PnD0H291XXh54Kl/8Nhv0aRNyTm1It7KQ7SRwtPM5edu+ZGiqaJxcnpTzK/kxI+Bts/KDq+oBQuOQBe/LAetxoq1R70JBOEm9g34MUcbSDhIhEGmNea57UlGpDCvfBN09VH2ncyxfG3gUX/wmCujT52zrKK/AWwctLW2Oq9WnIjbo/A4nAeyJy9OT1vU2fklJtSPEhWPlPiB0Jq1+oWpzOvB7uSbYHdW2G4gTw6qqtXPXsN3y2IRdPuedRqfpqSAvKGGNeEpEi4CMRuQHQP8uUqonjCPywEFb9G47sr7qt7/n2jLZ9z23WFApKyngpcQsHihzc9cYanp16NpNH9m7W91SqKTWkQB0AMMa87ixSHwPaT1WpyspK4afX7fuZDu+quq3bEPtG2yHXtEgHiDe+28aBIrvF1qdzIBPP7Nns76lUU6rPfVBibOOPrjPGvCsixcCiyvs0X5pKebiyUli3xC5MB7dV3RZ6Klz6VzjrppMaAeJkFDvK+c+qYwPC3nvZafj5eNLQm0rVrT4tqJUi8h7woTFme6X1nwM3i8hiYCXOYqVUu+Iohp/egFXPwKGcqtuCe9pTYJx9G/j4tWha76Rkk1dQCkCvTgFcf3b1GXWV8nT1KVATgTuAt53DEB0EArE7WHwOPGOM+an5UlTKA5UWQsprkPQsFOyuui2wM1x0P5z7e/Br+bPgjvIKFiQeaz1F/Wqgtp5Uq1Sf+6CKgfnAfGfvva7AEWPMweZOTimPU7Qfkv8D378IRfuqbgvqChfcY9/L1Igx8xrro9SdVSYjDD+3n9tyUaoxGjpYrAAHjTFHmiMZpTxW3ma7m3jq21B23Nc/uCdcOB1G/9YtLabKKioMLyZuccWRF/Qn0K9lrnsp1dTqXaBEZDowGygWkUPAC8aY55stM6XczRjYlgTfPQ8Zn1JlinWATn3tEcZHRYBvgFtSPN6K9N1s3lMAQLC/D7eN6+/ehJRqhPr04osFfgSmA0ONMXtEpBvwNxGZa4x5pLmTVKpFlRTA+v9CyquQu6769p5nwbh77RttW7jzQ10WVGo9TTu/nw4Iq1q1evXiA87BvvaU5Gw9rQPWA38QkaeNMQdO9AJKtQq7N9hFae1SKD1cffvgK+1rTP0v9tiBXP998ygWfJ3J8rU7+d1FA9ydjlKNUu8p30Xkc+B+YBcwAhiJfcpvO9DRGDOouZJsLB3NXNWq5LA9c+2axZC9uvp2nwAYcTOMi4ZurWeCv6LSMoL8PGnCbKVq1xRTvkcD7wCp2K2nocB6Y4wlIp51nkOpE6koh61fw9q37eLkKKq+T9hgGHMHjJpqdxtvZbQ4qbag3t9iY8wmERkLXI7deloHPODcVto86SnVRIyB3Wmw/l1Y9w4c3ll9Hy8fGHqtXZg8+DSeUu1Fg/7Mchaij50PpTybMbBrrT330sYP7bmYatJtqN1SGhEOIT1aNscmkH/EwetJWUwd24+uwTqXlGo79DyAalvKyyDnB7tb+MYPq4+Ld1RQV3tsvJHh0Gtkq24tLflhO0+t+IXnVm7mvstO457LBrs7JaWahBYo1foV5sGmFbDpc9jyBRTn17yfXwicMRGGT4HTJoB36++CXVZeweKkLABKyyroHuIZ92Mp1RS0QKnWp7QIsr+HrG8g8yvY8SPVbqI9yr8jnHEVDLsOBl3mMTfUNpX4DbnszC8GIKyDH5NH6XxPqu1oyEgS/sAUoH/l44wxjzV9WkpV4jgCOcmw9Ru7KOWkVJ2Z9ngdT4HBl9uFaaAFPm33usx/Vm11LU87/1QCfHVYI9V2NKQF9SGQD6wBSponHdXuGWN3ZshJsYvSjhTIXQ8VZbUfI97Qd6xdlAZfAT3ObNXXlOrrx+0H+Gm7PWazn7cXt55/qpszUqppNaRA9THGTGy2TFT7U1EO+7fC7vWQm2b3uNuRAkfqMTBJtyF2V/ABF8OAX7XKe5Ua69VKrafJo3rTLaTtthRV+9SQApUkImcZY9Y3WzaqbTIGDufCvk2wN8NuEe1Ogz3pNd8kW5OwwdD/Irsg9b8Ygrs3b84ebsfBI3yaluuK77hQhzVSbU9DCtRFwG9FZCv2KT4BjDFmRLNkploXY6D4IBzcDvs229NT7NsEeZtg35aax7arTUAo9BkDfc61n08Z3S5bSCfyelIW5RV2x5BxA8MY1rujmzNSquk1pEBNarYslOcrL4PCPXB4FxzMtgtRvvP5aNyQInRUh+7Qczj0GG6PEt77HAgb1C6uIZ2swpIy3v5huyvWQWFVW9WQoY62ichI4GLnqm+MMWubJy3VIspK7Os9RfvhyH4o3AuHd9tTmBfstk/LHX0u2ketXbnrw78TdD3NPlXXY9ixgtTOT9WdjG37iugU5Muh4jL6hwVx2RD9DFXb1NAJC38PLHOuihORl40xzzVLZurEjLGv35QU2CNylx62n0sKoLQASg4d23bkgF2AjhaiIwftZUdh0+bkG2RP4tdlAISdBl0H2wWp62Do0E1bRU1kWO+OfDXzUlZs3I0IeHnp56rapoac4vsdMNYYUwggIk8A3wFNVqBEZCIQC3gDrxhjYo7b7g+8DowG9gE3G2Oymur9662iwr4Pp9wB5aV2F+hal0vt2LW/w265lB0BR3HDn48WpdLDYCpa9ucO6gohPe0iFNoXQvs5l/vZj6AwLUItxNtLmDi8p7vTUKpZNaRACVBeKS53rmsSIuINvIA9WnoOkCwiHxljNlba7XfAAWPMaSISDjwB3NxUOVTz9Tz4YWH1YmPK6z62NfDysTsfBHaxn4PC7MFSg3seew7ubhelDt3axNBASqnWw6sB+74GfC8ij4rI34DvgVebMJfzgM3GmEznqOlLgOuO2+c6YLFz+V1gvEjdf7JnZGSwaNEiABwOB5ZlERcXB0BRURGWZbF06VIA8vPzsSyLZcuW2a2Vglz7tFjJISgr9qzi5BNIiW8ncor8KQ0bCqdeyJ7Qc0jY3ZmiYTfDBffyc68pxG7qw6Er/g0Ry/jytNmErx5G/t0b4JE8lvb7O9a7QRRNfR+mvkXcwdFYj36KY8Q0OGMii1asw5p8i6s4LVy4kAkTJrhSmD9/PpMmHes/Exsby+TJk13xvHnzmDJliiuOiYkhPDzcFc+dO5eIiAhXPHv2bCIjI13xrFmziIqKcsUzZ84kOjraFc+YMYMZM2a44ujoaGbOnOmKo6KimDVrliuOjIxk9uzZrjgiIoK5c+e64vDwcGJijjXcp0yZwrx581zx5MmTiY2NdcWTJk1i/vz5rnjChAksXLjQFVuWdXLfPSAvLw/Lsli+fDkA2Tt2YlkW8fHxdpydjWVZJCQkAJCZmYllWSQmJgL2996yLJKSkgBIS0vDsiySk5MBSE1NxbIsUlNTAUhOTsayLNLS0gBISkrCsiwyMjIASExMxLIsMjPtUeETEhKwLIvs7GwA4uPjsSyL3Fy7+/vy5cuxLIu8vDwAli1bhmVZ5OfbYyUuXboUy7IoKrJvNYiLi8OyLBwOe5SQRYsWYVmW67PU7577vnu5ubnN+t2rTb0LlDHmaSAS+9RaHnC7Mebf9T2+Hk4BsivFOc51Ne5jjCnDHtkirKYXE5EoEUkRkZSjX/gG86q9xVAh3uAbRIVfRw6U+lDs1wVC+1Ea0o+thQEcDh4Avc+hqOtIfjoQzMGws2HwlRzq/Su+3BPK3j5Xwrl3sue0m3k9qwc7h94JV/6DrOHT+Wd6P7aPfQymLmHD6H9wz4+DybrydYj+gdXjXuGaVWeRGZEMD+fyzdhXifhhGLt//V+I/IQfh/yFx9P7c+hXj8EVj7Op9/W8v6MbpadPhtPGczBoALnF/uAfoqfjWpknvtjOrmFT+WmvcXUxV6otq3PKdxFZZYy5SEQOY3fjqvxbzRhjmuQGDBG5EZhojLnTGd+Kfc3rnkr7pDn3yXHGW5z75J3otU96yvfifCgtBG8/+3SYt5/dkvDy0V/uqkXtPVzChTFfUlpuX3d8/+4LOLuf3hum2oaTnvLdGHOR8zmkORKrZAfQt1Lcx7mupn1yRMQH6ITdomseAZ3sh1Ju9ub321zFaVTfUC1Oql2o9yk+Z6+9Otc1QjIwWEQGiIgfEA58dNw+HwG3O5dvBL40dTUBlWrlih3lxK0+NvGi3pir2ouGdJK4vIZ1TTa6hPOa0j3AZ0A68I4xZoOIPCYiR698/gcIE5HNwP8BDzbV+yvlqT5au5O8glIAenUK0O7lqt2o8xSfiPwRuBsYKCLrKuO9iZEAABxXSURBVG0KAb5tymSMMZ8Anxy3bnal5WLgpqZ8T6U8mTGmyqjlt1/QH1/vhvxdqVTrVZ/7oN4CPgX+SdUWy2FjzP5myUopBUDSln38nGuPcRjo683Uc/u5OSOlWk59OknkY3fnntr86SilKnvlm0zX8m/G9KFTkN4srdqPhowkgYh0BgYDAUfXGWO+buqklFKweU8BKzP2AvZdDZE655NqZxoyWOydwHTs7t+pwPnYY/Fd1jypKdW+vfbtsWtP44f0oH/XDm7MRqmW15CrrdOBc4FtxphLgbOBg82SlVKK28b15+YxffHz8eLOi7X1pNqfhpziKzbGFIsIIuJvjPlZRM5otsyUaufO6BnCEzeO4C+ThtBZrz2pdqghBSpHREKBD4AVInIA2FbHMUqpRurSwc/dKSjlFg0pUKuADsaYR0VkJfYwQ/HNk5ZSSqn2riHXoIKBz0XkG+As4HvntBhKqSZijOGt77dzsEj/aynVkOk2/maMOROIBnoBiSKS0GyZKdUOrc7cz1/fX8/5//yCx/+3se4DlGrDTmbMlD1ALvYo4t2bNh2l2rcFX28BoNhRQWGpB02OqZQbNGQ087tF5CvgC+xJAn9vjBnRXIkp1d5s3HmIryrdmBv1q4Fuzkgp92pIJ4m+wAxjzInn6FVKnZSjrSeAScN7MkBvzFXtXL0LlDFmVnMmolR7lr2/iOVrd7riP1wyyI3ZKOUZ6jPdxvFTvrs20YRTvivVni38JpMK5/+uC08LY0SfUPcmpJQH8KQp35Vql/IKSlianO2K/3jJaW7MRinP4UlTvivVLi1OyqKkrAKA4ad05MLTwtyckVKewWOmfFeqPSooKeP1746NGPbHS05DRNyYkVKew6OmfFeqvcneX0RokC/5RxycGhbExOE93Z2SUh5Dp3xXyo2G9urIF/93CR+k7iTY3wdvL209KXWUTvmulJv5eHtx4+g+7k5DKY/TkE4Si53TbRyNO4vIq82TllJKqfauIZ0kRhhjXDPoGmMOYM+qq5RqoJ0Hj2CMqXtHpdqxhhQoLxHpfDQQkS40bKgkpRR2z72rn/2G3yz4jm8352mhUqoWDSlQTwHfichcEXkcSAL+1TxpKdV2Lfp2KweKHCRnHeAv762jrEILlFI1ach8UK8DNwC7gV1AFHB+M+WlVJu0v7CUlxIzXfG9l52Gr/fJzHqjVNvX0FN0/sCpwE3AVuC9Js9IqTbs+S83U1BSBsCgbh2Yco723lOqNvW5Ufd07C7mU4E8YCkgxphLmzk3pdqU7P1FvLE6yxU/MHEIPtp6UqpW9WlB/Qx8A1xjjNkMICL3N2tWSrVBT6/4BUe5fb3pnH6hXDGsh5szUsqz1efPtxuwrzmtFJGFIjIee6qNJiMiXURkhYhscj53rmW/chFJdT4+asoclGpOaTvy+SB1hyueddVQHXNPqTrUWaCMMR8YY8KBIcBKYAbQXUReFJErmiiPB4EvjDGDsaeUf7CW/Y4YY0Y5H5Ob6L2ValbGGB5bvpGjvcknDO3Buf27uDcppVqBhvTiKzTGvGWMuRboA/wE/KWJ8rgOWOxcXgz8uoleF4CMjAwWLVoEgMPhwLIs4uLiACgqKsKyLJYuXQpAfn4+lmWxbNkyAPLy8rAsi+XLlwOQm5uLZVnEx8cDkJ2djWVZJCQkAJCZmYllWSQmJrre27IskpKSAEhLS8OyLJKTkwFITU3FsixSU1MBSE5OxrIs0tLSAEhKSsKyLDIyMgBITEzEsiwyM+2eYAkJCViWRXa2PZ9QfHw8lmWRm5sLwPLly7Esi7y8PACWLVuGZVnk5+cDsHTpUizLoqioCIC4uDgsy8LhcACwaNEiLMtyfZYLFy5kwoQJrnj+/PlMmnRsUPvY2FgmTz72t8O8efOYMmWKK46JiSE8PNwVz507l4iICFc8e/ZsIiMjXfGsWbOIiopyxTNnziQ6OtoVz5gxgxkzZrji6OhoZs6c6YqjoqKYNevYZNCRkZHMnj3bFUdERDB37lxXHB4eTkxMjCueMmUK8+bNc8WTJ08mNjbWFU+aNIn58+e74gkTJrBw4UJXbFkWM59bwg9ZzmErK8o5s+wXQL97+t1r/u9ea/m9V5uTukJrjDlgjHnZGDP+ZI6vQQ9jzC7nci5Q28n5ABFJEZHVInLCIiYiUc59U45+4ZVyh4MOL3ycg8B2zP2Rbv4Vbs5IqdZBWuoudhFJAGqaS+AhYLExpvI4fweMMdWuQ4nIKcaYHSIyEPgSGG+M2VLXe48ZM8akpKQ0InulGmfT7sM8++Vm/n79cDoG+Lo7HaU8ioisMcaMOX59iw1VZIyZUNs2EdktIr2MMbtEpBewp5bX2OF8zhSRr7DHAqyzQCnlboN7hPDcVB26UqmG8JSbMD4Cbncu3w58ePwOztHT/Z3LXYELgY0tlqFSSqkW5SkFKga4XEQ2AROcMSIyRkRece4zFEgRkbXYvQljjDFaoJRHWpmxh+z9Re5OQ6lWzSNGIzfG7AOqdbgwxqQAdzqXk4CzWjg1pRps96Fi7nv7J8rKDX+64nQiLxygM+UqdRI8pQWlVJtgjOGRD9I4XFzGEUc5cau34SjXXntKnQwtUEo1of+uyeHzjbtd8T9uOIsAX283ZqRU66UFSqkmkpVXyN8+2uCKp43txwWDuroxI6VaNy1QSjUBR3kFM5amUlhaDsDAbh14+Ophbs5KqdZNC5RSTeDpFb+Qmn0QAB8vIfbmswn001N7SjWGFiilGik+LZcXvzp2v/j/XXE6Z/Xp5MaMlGobtEAp1Qib9xQw879rXfElp3fjrl8NcmNGSrUdWqCUaoTU7IMUldpTuPftEkhs+Ci950mpJuIRN+oq1VrdOLoPvUMD+PN/1/HitNGEBvm5OyWl2gwtUEo10gWDurJypoWfj56QUKop6f8opRrAGMOh4urzi2lxUqrp6f8qpRrgmYRNXPPsKnYcPOLuVJRq87RAKVVPz36xidgvNrF9fxE3L/iOXflapJRqTnoNSqk6GGN48rMM5le612lQt2A6a4cIpZqVFiilTsBRXsGD763nvR9zXOsuHtyVBbeO1kFglWpmWqCUqkV+kYN7l/zE17/sda0bP6Q7z99yjhYnpVqAFiilapC+6xB3vbGG7ZVmxb15TF/+fv1wfLz10q1SLUELlFLHeXdNDg9/sJ5ix7GJBu8bP5j7JwxGREeJUKqlaIFS6jh5BSWu4tTBz5unfjOSicN7uTkrpdofLVBKHef3Fw/ki/Td7C8sZcGtozmte4i7U1KqXdICpdq1tB35iMCZvY9Nj+HtJbxwyzmEBPjqnE5KuZEWKNUubd5zmGcSNvHx+l0M792JD6IvrDIKefeOAW7MTikFWqBUO7M2+yD/WbWV5et2Yoy9bv2OfJYkb2fa2FPdm5xSqgotUKrNK3aU89mGXBYnZfHj9oPVtk8Y2oNxA8PckJlS6kS0QKk2KzX7IG99v41P1+dyuKSs2vZLz+jG/Zefzog+oW7ITilVFy1Qqs1KydrPOyk5Vdb5eXtxzche/PaC/lqYlPJwWqBUq7Ur/wgpWQdYs+0AaTvyWRJ1fpVRHq4d2Zu/f5KOMXBqWBBTzunD1PP60S3E341ZK6XqSwuU8niHix1s21fEz7mH+XnXIfs59xB5BaVV9vs59zDDTznWXbxHxwDmXDOMkX1DGdU3VEeBUKqV0QKl3KqotIy8w6XsLSime0gAfbsEVdl+/fxv+amGjg01WbPtQJUCBfDbCwc0Wa5KqZblEQVKRG4CHgWGAucZY1Jq2W8iEAt4A68YY2JaLEmFMYaSsgpKHBUUl5VT7CinpKyCYkc5xY4KQoN8Ob1H1VEXVmzczerMfRw64uBwcRmHiu3n/CMO9hWUUFha7tr3z1eeQfSlp1U5vlOgb635BPp6M7JvJ87t34XRp3bmnFM7N+0PrJRyK48oUEAacAOwoLYdRMQbeAG4HMgBkkXkI2PMxuZMrKy8gofeT8NgXPfNGMAYMBxb4VzisevOJCTg2C/V/YWlzPloA8a49nbuX+n1nLG3lzB/2ugq779lbwH/dF5Hsd/XVMnh2LIhrIMfz4SfXeX4H7bu55mEX6gwhooKKKuooLzCUFZhKK/0KKswnNm7Iy9GVH3/N77L4vGP0137nMg1I3rx/C3nVFmXtCWP177NOuFxR+09XFJtXb8uQfj5eNG3cyCDu4dwRs8QhvYKYUjPjvTrEoSXl562U6qt8oh5A4wx6caYjDp2Ow/YbIzJNMaUAkuA6+rz+hkZGSxatAgAh8OBZVnExcUBUFRUhGVZLF26FID8/Hwsy2LZsmUA5OXlsTQlm3dScvjvGvvx7poc3vsxh2U/7rAfP+3gfefjly1bsSyLxMREADb8/AvL1+7kf+t28fHRx/pdfLI+l0/T7Ef8hlw+27CbzzbsJjk5GcuySEtLA2DV9z+SkL6HL37ew5c/72Flxl5WZuzlq4y9JP5iP77+ZS/fbMrj+637iY+Px7IscnNzAfgs8VuStuxjdeZ+fsjaz4/bD7I2J58NO+1rOZv2FJCZV8j2/UXsyi8mLi4Oy7JwOBwAfLd6NSVlFXUWJ4CSsgpiY2OZPHmya13aj8knPMaLCvzLChnZpxM9OwUwe/ZsIiMjXduPfPc2l+x5ny/+ZPHSraPZ8dnLLH/xcfp37YCXlzBjxgxmzJjh2j86OpqZM2e64qioKGbNmuWKIyMjmT17tiuOiIhg7ty5rjg8PJyYmGMN8ylTpjBv3jxXPHnyZGJjY13xpEmTmD9/viueMGECCxcudMWWZTXqu2dZFsuXLwcgNzcXy7KIj48HIDs7G8uySEhIACAzM7PKdy8jIwPLskhKSgIgLS0Ny7JITrb/TVJTU7Esi9TUVIBq372kpCQsyyIjw/6vmZiYiGVZZGZmApCQkIBlWWRnZwNU++4tX74cy7LIy8sDYNmyZViWRX5+PgBLly7FsiyKiuwpTY7/7i1atAjLslyf5cKFC5kwYYIrnj9/PpMmTXLFx3/35s2bx5QpU1xxTEwM4eHhrnju3LlERES44uO/e7NmzSIqKsoVz5w5k+joaFes372m++7VxlNaUPVxCpBdKc4Bxta2s4hEAVEA/v6N6LXl5gvrjX37hhxeYaoXoeOP9xaDcZQQ1rkTAb5eFB3Kp/DQQYYPPYNB3YLhQNX9+/kVsmv7Vu6563eEBPjyyYfvsW1TOrFPPUHXDv7Ezovhl18yiJtn/8eZ/UXV432kokE/g1Lu4HA4uPLKKwkLCyM9PR2A2267DS8vL1d8xx134OPj44rvuuuuKvHdd9+Nn5+fK54+fTr+/v6u+E9/+hMBAQGu+MEHHyQwMNAVP/zwwwQFBbniOXPmEBwcTHp6OsaYKnFFRQVz5swhJCSkStyxY0fS09MpLy+vMe7UqRPp6emUlZUxZ84cQkNDq8RH93c4HMyZM8eVT3l5ObNnz8bHx8f1B0h9iKnhl1JzEJEEoGcNmx4yxnzo3OcrYGZN16BE5EZgojHmTmd8KzDWGHNPXe89ZswYk5JS42WtOlVUGN5JyXbmAHL016XYv7yP9gwT5/arzupVZbbVI6XlfL4x9+jP4Pple/S15NjLISJMHF71I8ovcvBD1v4q73HsGKFSOvj7eDNuUNUREfIKSsjIPYxgD4Lq4y14e3nhLVIpFrxFCPD1pmenqmPQOcrtU4I+XvZ+2hNOqeq2bt1KSEgIYWFh+n+kFsYY9u3bx+HDhxkwoGrnJRFZY4wZc/wxLdaCMsZMqHuvE9oB9K0U93Gua1ZeXkL4ef1O+vhAP2+uG3XKSR/fKciXy4f1OOnjuwb70/W0k29B+np7obObK3VixcXF9O/fX4vTCYgIYWFh7N27t97HeMQ1qHpKBgaLyAAR8QPCgY/cnJNSSgFocaqHhn5GHlGgROR6EckBxgEfi8hnzvW9ReQTAGNMGXAP8BmQDrxjjNngrpyVUko1L4/oJGGMeR94v4b1O4GrKsWfAJ+0YGpKKeXx9u3bx/jx4wG7x523tzfdunUD4IcffsDPz6/erxUREcGNN97Ir3/962bJtSE8okAppZQ6eWFhYa4u248++ijBwcFVury3VlqglFKqif17xS/EfrGpXvtOPa8v/7xhRJV1s5at4+0fjt1VM338YO6//PSTyuXaa69l586dFBcXc//993PnnXdSVlZG165d+cMf/sCnn35KUFAQH374Id27dwdg5cqV/Otf/yI3N5ennnqK66+//qTeu7E84hqUUkqp5rF48WLWrFlDcnIyTz/9NAcO2Dcr5ufnc8kll7B27VrGjRvHq6++6jpmz549fPvtt3zwwQdVbjZuaVqglFKqDfv3v//NyJEjGTduHDk5OWzZsgWAwMBA10gco0ePJisry3XMr3/9a0SEESNGsGNHs9/NUys9xaeUUk3s/stPP+lTcgD/vGFEtdN+JyMhIYGvv/6a1atXExgYyEUXXURxcTFAlY4T3t7elJUdm3W68ug7LTWYQ020BaWUUm1Ufn4+Xbp0ITAwkA0bNrjGwmsttEAppVQbdfXVV1NUVMSwYcN4+OGHGTu21uFLPVKLjcXnTo0Zi08ppeqSnp7O0KFD3Z1Gq1DTZ1XbWHzaglJKKeWRtEAppZTySFqglFJKeSQtUEoppTySFiillFIeSQuUUkopj6QFSiml2oDg4OBGv8bOnTu58cYba91+8OBB5s+fX+/9G0sLlFJKKQB69+7Nu+++W+v24wtUXfs3lo7Fp5RSTenRTs342vkN2j0rK4s77riDvLw8unXrxmuvvUa/fv3YsmUL06ZNo7CwkOuuu45nnnmGgoICsrKyuOaaa0hLS2PDhg1ERkZSWlpKRUUF7733Ho888ghbtmxh1KhRXH755URHR7v2Ly8v5y9/+Qvx8fF4eXnx+9//nnvvvbdRP662oJRSqo269957uf3221m3bh3Tpk3jvvvuA2D69OlMnz6d9evX06dPnxqPfemll5g+fTqpqamkpKTQp08fYmJiGDRoEKmpqTz55JNV9n/55ZfJysoiNTXV9X6NpQVKKaXaqO+++45bbrkFgFtvvZVVq1a51t90000Aru3HGzduHP/4xz944okn2LZtG4GBgSd8r4SEBO666y58fOwTc126dGl0/nqKTymlmlIDT8N5qltuuYWxY8fy8ccfc9VVV7FgwQIGDhzYojloC0oppdqoCy64gCVLlgDw5ptvcvHFFwNw/vnn89577wG4th8vMzOTgQMHct9993Hdddexbt06QkJCOHz4cI37X3755SxYsMA1r9T+/fsbnb8WKKWUagOKioro06eP6/H000/z3HPP8dprrzFixAjeeOMNYmNjAXjmmWd4+umnGTFiBJs3b6ZTp+odO9555x2GDx/OqFGjSEtL47bbbiMsLIwLL7yQ4cOH8+c//7nK/nfeeSf9+vVjxIgRjBw5krfeeqvRP5NOt6GUUo3U2qbbKCoqIjAwEBFhyZIlvP3223z44Yct8t4NmW5Dr0EppVQ7s2bNGu655x6MMYSGhvLqq6+6O6UaaYFSSql25uKLL2bt2rXuTqNOeg1KKaWaQHu4XNJYDf2MtEAppVQjBQQEsG/fPi1SJ2CMYd++fQQEBNT7GD3Fp5RSjdSnTx9ycnLYu3evu1PxaAEBAbWOXFETLVBKKdVIvr6+DBgwwN1ptDkecYpPRG4SkQ0iUiEi1boaVtovS0TWi0iqiGi/caWUasM8pQWVBtwALKjHvpcaY/KaOR+llFJu5hEFyhiTDiAi7k5FKaWUh/CIAtUABvhcRAywwBjzcm07ikgUEOUMC0QkoyUSbEZdgfbectTPwKafg00/B1tb+BxOrWllixUoEUkAetaw6SFjTH3H2LjIGLNDRLoDK0TkZ2PM1zXt6CxetRaw1kZEUmoaCqQ90c/App+DTT8HW1v+HFqsQBljJjTBa+xwPu8RkfeB84AaC5RSSqnWzSN68dWHiHQQkZCjy8AV2J0rlFJKtUEeUaBE5HoRyQHGAR+LyGfO9b1F5BPnbj2AVSKyFvgB+NgYE++ejN2izZyubAT9DGz6Odj0c7C12c+hXUy3oZRSqvXxiBaUUkopdTwtUEoppTySFqhWSET+JCJGRLq6O5eWJiJPisjPIrJORN4XkVB359SSRGSiiGSIyGYRedDd+bQ0EekrIitFZKNzeLTp7s7JnUTEW0R+EpH/uTuX5qAFqpURkb7YPRi3uzsXN1kBDDfGjAB+AWa5OZ8WIyLewAvAJGAYMFVEhrk3qxZXBvzJGDMMOB+IboefQWXTgXR3J9FctEC1Pv8GHsAeVaPdMcZ8bowpc4argfqP3d/6nQdsNsZkGmNKgSXAdW7OqUUZY3YZY350Lh/G/uV8inuzcg8R6QNcDbzi7lyaixaoVkRErgN2GGM8f67mlnEH8Km7k2hBpwDZleIc2ukvZwAR6Q+cDXzv3kzc5hnsP1Yr3J1Ic2ltY/G1eScaEgr4K/bpvTatPsNiichD2Kd73mzJ3JRnEJFg4D1ghjHmkLvzaWkicg2wxxizRkQsd+fTXLRAeZjahoQSkbOAAcBa56jvfYAfReQ8Y0xuC6bY7OoaFktEfgtcA4w37etGvh1A30pxH+e6dkVEfLGL05vGmGXuzsdNLgQmi8hVQADQUUTijDERbs6rSemNuq2UiGQBY9rb3FgiMhF4GrjEGNOu5tcWER/sjiHjsQtTMnCLMWaDWxNrQWL/dbYY2G+MmeHufDyBswU10xhzjbtzaWp6DUq1Ns8DIdij2aeKyEvuTqilODuH3AN8ht054J32VJycLgRuBS5z/vunOlsRqg3SFpRSSimPpC0opZRSHkkLlFJKKY+kBUoppZRH0gKllFLKI2mBUkop5ZG0QCmllPJIWqCUUkp5JC1QSjUR5xxdT1WKZ4rIoyfYP1RE7j7B9qQmTrHO91TKk2iBUqrplAA3NGAiyVCg1mJhjLmgSbJqwHsq5Um0QCnVdMqAl4H7j98gIhEi8oNzaJ4FzskHY4BBznVP1nBMgfO5v4iki8hC5yyyn4tIoHP9zyLypnP7uyISVOmYtEqvdbQ1V9d7fllpCKFiEflNE302SjWYFiilmtYLwDQR6XR0hYgMBW4GLjTGjALKgWnAg8AWY8woY8yf63jdwcALxpgzgYPAFOf6M4D5xpihwCHqbh2d8D2NMZc5c1wAfIQ9arhSbqEFSqkm5Jyb6HXgvkqrxwOjgWQRSXXGAxv40luNManO5TVAf+dytjHmW+dyHHDRyeRdmYjchj2t/DRjTHljX0+pk6XzQSnV9J4BfgRec8YCLDbGzKq8k3NG2PoqqbRcDgQ6l48f7floXEbVP0AD6vMmInITduvuOmOMowH5KdXktAWlVBMzxuwH3gF+51z1BXCjiHQHEJEuInIqcBh76pDG6Cci45zLtwCrnMu7ge4iEiYi/tgTPHKi93TO0no3cIMxpriReSnVaFqglGoeTwFdAYwxG4GHgc9FZB2wAuhljNkHfCsiaTV1WKinDCBaRNKBzsCLzvd0AI8BPzjf72fn+hO952LsWXq/dXaS+B1KuZHOB6VUK+U8Rfg/Y8xwN6eiVLPQFpRSSimPpC0opZRSHklbUEoppTySFiillFIeSQuUUkopj6QFSimllEfSAqWUUsojaYFSSinlkbRAKaWU8kj/DxHZXRdxojQEAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "def tanh(z):\n",
    "    e_p = np.exp(z)\n",
    "    e_m = np.exp(-z)\n",
    "    return (e_p - e_m) / (e_p + e_m)\n",
    "\n",
    "z = np.arange(-5, 5, 0.005)\n",
    "log_act = logistic(z)\n",
    "tanh_act = tanh(z)\n",
    "plt.ylim([-1.5, 1.5])\n",
    "plt.xlabel('Net input $z$')\n",
    "plt.ylabel('Activation $\\phi(z)$')\n",
    "plt.axhline(1, color='black', linestyle=':')\n",
    "plt.axhline(0.5, color='black', linestyle=':')\n",
    "plt.axhline(0, color='black', linestyle=':')\n",
    "plt.axhline(-0.5, color='black', linestyle=':')\n",
    "plt.axhline(-1, color='black', linestyle=':')\n",
    "plt.plot(z, tanh_act,\n",
    "    linewidth=3, linestyle='--',\n",
    "    label='Tanh')\n",
    "plt.plot(z, log_act,\n",
    "    linewidth=3,\n",
    "    label='Logistic')\n",
    "plt.legend(loc='lower right')\n",
    "plt.tight_layout()\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([-0.9999092 , -0.99990829, -0.99990737, ...,  0.99990644,\n",
       "        0.99990737,  0.99990829])"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.tanh(z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=134668, shape=(2000,), dtype=float64, numpy=\n",
       "array([-0.9999092 , -0.99990829, -0.99990737, ...,  0.99990644,\n",
       "        0.99990737,  0.99990829])>"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "tf.keras.activations.tanh(z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.00669285, 0.00672617, 0.00675966, ..., 0.99320669, 0.99324034,\n",
       "       0.99327383])"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from scipy.special import expit\n",
    "\n",
    "expit(z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=134670, shape=(2000,), dtype=float64, numpy=\n",
       "array([0.00669285, 0.00672617, 0.00675966, ..., 0.99320669, 0.99324034,\n",
       "       0.99327383])>"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.keras.activations.sigmoid(z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Rectified linear unit activation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=134672, shape=(2000,), dtype=float64, numpy=array([0.   , 0.   , 0.   , ..., 4.985, 4.99 , 4.995])>"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "tf.keras.activations.relu(z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Summary"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Appendix\n",
    "\n",
    "## Splitting a dataset: danger of mixing train/test examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0, 1, 2, 3, 6, 7, 9, 10, 11, 13} {4, 5, 8, 12, 14}\n"
     ]
    }
   ],
   "source": [
    "## the correct way:\n",
    "ds = tf.data.Dataset.range(15)\n",
    "ds = ds.shuffle(15, reshuffle_each_iteration=False)\n",
    "\n",
    "\n",
    "ds_train = ds.take(10)\n",
    "ds_test = ds.skip(10)\n",
    "\n",
    "ds_train = ds_train.shuffle(10).repeat(10)\n",
    "ds_test = ds_test.shuffle(5)\n",
    "ds_test = ds_test.repeat(10)\n",
    "\n",
    "set_train = set()\n",
    "for i,item in enumerate(ds_train):\n",
    "    set_train.add(item.numpy())\n",
    "\n",
    "set_test = set()\n",
    "for i,item in enumerate(ds_test):\n",
    "    set_test.add(item.numpy())\n",
    "\n",
    "print(set_train, set_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14} {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14}\n"
     ]
    }
   ],
   "source": [
    "## The wrong way:\n",
    "ds = tf.data.Dataset.range(15)\n",
    "ds = ds.shuffle(15, reshuffle_each_iteration=True)\n",
    "\n",
    "\n",
    "ds_train = ds.take(10)\n",
    "ds_test = ds.skip(10)\n",
    "\n",
    "ds_train = ds_train.shuffle(10).repeat(10)\n",
    "ds_test = ds_test.shuffle(5)\n",
    "ds_test = ds_test.repeat(10)\n",
    "\n",
    "set_train = set()\n",
    "for i,item in enumerate(ds_train):\n",
    "    set_train.add(item.numpy())\n",
    "\n",
    "set_test = set()\n",
    "for i,item in enumerate(ds_test):\n",
    "    set_test.add(item.numpy())\n",
    "\n",
    "print(set_train, set_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Splitting a dataset using `tfds.Split`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'features': <tf.Tensor: id=135238, shape=(4,), dtype=float32, numpy=array([6.1, 2.8, 4.7, 1.2], dtype=float32)>, 'label': <tf.Tensor: id=135239, shape=(), dtype=int64, numpy=1>}\n",
      "\n",
      "{'features': <tf.Tensor: id=135245, shape=(4,), dtype=float32, numpy=array([5.7, 3. , 4.2, 1.2], dtype=float32)>, 'label': <tf.Tensor: id=135246, shape=(), dtype=int64, numpy=1>}\n",
      "(<tf.Tensor: id=135290, shape=(4,), dtype=float32, numpy=array([6.8, 2.8, 4.8, 1.4], dtype=float32)>, <tf.Tensor: id=135291, shape=(), dtype=int64, numpy=1>)\n",
      "Training Set:  116 Test Set:  34\n",
      "10 12 12\n",
      "Training Set:  116 Test Set:  34\n",
      "10 12 12\n",
      "Training Set:  116 Test Set:  34\n",
      "10 12 12\n",
      "Training Set:  116 Test Set:  34\n",
      "10 12 12\n",
      "Training Set:  116 Test Set:  34\n",
      "10 12 12\n"
     ]
    }
   ],
   "source": [
    "\n",
    "##--------------------------- Attention ------------------------##\n",
    "##                                                              ##\n",
    "##     Note: currently, tfds.Split has a bug  in TF 2.0.0       ##\n",
    "##                                                              ##\n",
    "##  I.e., splitting [2, 1] is expected to result in             ##\n",
    "##      100 train and 50 test examples                          ##\n",
    "##                                                              ##\n",
    "##  but instead, it results in 116 train and 34 test examples   ##\n",
    "##                                                              ##\n",
    "##--------------------------------------------------------------##\n",
    "\n",
    "import tensorflow as tf\n",
    "import tensorflow_datasets as tfds\n",
    "import numpy as np\n",
    "\n",
    "##  method 1: specifying percentage:\n",
    "#first_67_percent = tfds.Split.TRAIN.subsplit(tfds.percent[:67])\n",
    "#last_33_percent = tfds.Split.TRAIN.subsplit(tfds.percent[-33:])\n",
    "\n",
    "#ds_train_orig = tfds.load('iris', split=first_67_percent)\n",
    "#ds_test = tfds.load('iris', split=last_33_percent)\n",
    "\n",
    "\n",
    "##  method 2: specifying the weights\n",
    "split_train, split_test = tfds.Split.TRAIN.subsplit([2, 1])\n",
    "\n",
    "ds_train_orig = tfds.load('iris', split=split_train)\n",
    "ds_test = tfds.load('iris', split=split_test)\n",
    "\n",
    "print(next(iter(ds_train_orig)))\n",
    "print()\n",
    "print(next(iter(ds_test)))\n",
    "\n",
    "\n",
    "ds_train_orig = ds_train_orig.shuffle(100, reshuffle_each_iteration=True)\n",
    "ds_test  = ds_test.shuffle(50, reshuffle_each_iteration=False)\n",
    "\n",
    "ds_train_orig = ds_train_orig.map(\n",
    "    lambda x: (x['features'], x['label']))\n",
    "\n",
    "ds_test = ds_test.map(\n",
    "    lambda x: (x['features'], x['label']))\n",
    "\n",
    "print(next(iter(ds_train_orig)))\n",
    "\n",
    "\n",
    "for j in range(5):\n",
    "    labels_train = []\n",
    "    for i,item in enumerate(ds_train_orig):\n",
    "        labels_train.append(item[1].numpy())\n",
    "\n",
    "    labels_test = []\n",
    "    for i,item in enumerate(ds_test):\n",
    "        labels_test.append(item[1].numpy())\n",
    "    print('Training Set: ',len(labels_train), 'Test Set: ', len(labels_test))\n",
    "\n",
    "    labels_test = np.array(labels_test)\n",
    "\n",
    "    print(np.sum(labels_test == 0), np.sum(labels_test == 1), np.sum(labels_test == 2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<br>\n",
    "<br>\n",
    "<br>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "Readers may ignore the next cell."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[NbConvertApp] Converting notebook ch13_part2.ipynb to script\n",
      "[NbConvertApp] Writing 13934 bytes to ch13_part2.py\n"
     ]
    }
   ],
   "source": [
    "! python ../.convert_notebook_to_script.py --input ch13_part2.ipynb --output ch13_part2.py"
   ]
  }
 ],
 "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": "3.7.4"
  },
  "toc-showmarkdowntxt": false,
  "toc-showtags": false
 },
 "nbformat": 4,
 "nbformat_minor": 4
}