notebooks/regression_in_tensorflow.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Regression Example in TensorFlow\n",
    "Predicting house prices in Boston, Massachusetts "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/jonkrohn/DLTFpT/blob/master/notebooks/regression_in_tensorflow.ipynb)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Load dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from tensorflow.keras.datasets import boston_housing # new!\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Dropout \n",
    "from tensorflow.keras.layers import BatchNormalization \n",
    "from tensorflow.keras.callbacks import ModelCheckpoint # new!\n",
    "import os # new!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "(X_train, y_train), (X_valid, y_valid) = boston_housing.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(404, 13)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(102, 13)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_valid.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  1.23247,   0.     ,   8.14   ,   0.     ,   0.538  ,   6.142  ,\n",
       "        91.7    ,   3.9769 ,   4.     , 307.     ,  21.     , 396.9    ,\n",
       "        18.72   ])"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "15.2"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Design neural network architecture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "\n",
    "model.add(Dense(32, input_dim=13, activation='relu'))\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Dense(16, activation='relu'))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Dropout(0.2))\n",
    "\n",
    "model.add(Dense(1, activation='linear'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense (Dense)                (None, 32)                448       \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 32)                128       \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 16)                528       \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 16)                64        \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 16)                0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 1)                 17        \n",
      "=================================================================\n",
      "Total params: 1,185\n",
      "Trainable params: 1,089\n",
      "Non-trainable params: 96\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Configure Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss='mean_squared_error', optimizer='adam', )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "output_dir = 'model_output/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "run_name = 'regression_baseline'\n",
    "output_path = output_dir + run_name"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "if not os.path.exists(output_path):\n",
    "    os.makedirs(output_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "modelcheckpoint = ModelCheckpoint(output_path + '/weights.{epoch:02d}.hdf5', # decimal integers\n",
    "                                  save_weights_only=True) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Train!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 404 samples, validate on 102 samples\n",
      "Epoch 1/32\n",
      "404/404 [==============================] - 1s 1ms/sample - loss: 574.9674 - val_loss: 670.9787\n",
      "Epoch 2/32\n",
      "404/404 [==============================] - 0s 283us/sample - loss: 560.1075 - val_loss: 610.3649\n",
      "Epoch 3/32\n",
      "404/404 [==============================] - 0s 272us/sample - loss: 538.9081 - val_loss: 569.0831\n",
      "Epoch 4/32\n",
      "404/404 [==============================] - 0s 292us/sample - loss: 523.4382 - val_loss: 532.6724\n",
      "Epoch 5/32\n",
      "404/404 [==============================] - 0s 346us/sample - loss: 498.2580 - val_loss: 513.4582\n",
      "Epoch 6/32\n",
      "404/404 [==============================] - 0s 382us/sample - loss: 473.1501 - val_loss: 482.6169\n",
      "Epoch 7/32\n",
      "404/404 [==============================] - 0s 349us/sample - loss: 444.2933 - val_loss: 448.3360\n",
      "Epoch 8/32\n",
      "404/404 [==============================] - 0s 316us/sample - loss: 408.2028 - val_loss: 410.1629\n",
      "Epoch 9/32\n",
      "404/404 [==============================] - 0s 311us/sample - loss: 374.2431 - val_loss: 365.4785\n",
      "Epoch 10/32\n",
      "404/404 [==============================] - 0s 330us/sample - loss: 333.5626 - val_loss: 320.0330\n",
      "Epoch 11/32\n",
      "404/404 [==============================] - 0s 289us/sample - loss: 297.5275 - val_loss: 190.4927\n",
      "Epoch 12/32\n",
      "404/404 [==============================] - 0s 296us/sample - loss: 246.6466 - val_loss: 157.5952\n",
      "Epoch 13/32\n",
      "404/404 [==============================] - 0s 335us/sample - loss: 210.3841 - val_loss: 130.2611\n",
      "Epoch 14/32\n",
      "404/404 [==============================] - 0s 355us/sample - loss: 162.7810 - val_loss: 117.6108\n",
      "Epoch 15/32\n",
      "404/404 [==============================] - 0s 327us/sample - loss: 144.0858 - val_loss: 127.0299\n",
      "Epoch 16/32\n",
      "404/404 [==============================] - 0s 303us/sample - loss: 117.4293 - val_loss: 140.5821\n",
      "Epoch 17/32\n",
      "404/404 [==============================] - 0s 335us/sample - loss: 93.0745 - val_loss: 83.5203\n",
      "Epoch 18/32\n",
      "404/404 [==============================] - 0s 360us/sample - loss: 81.5223 - val_loss: 131.9906\n",
      "Epoch 19/32\n",
      "404/404 [==============================] - 0s 341us/sample - loss: 64.4891 - val_loss: 100.1118\n",
      "Epoch 20/32\n",
      "404/404 [==============================] - 0s 353us/sample - loss: 67.9189 - val_loss: 30.7400\n",
      "Epoch 21/32\n",
      "404/404 [==============================] - 0s 345us/sample - loss: 51.6477 - val_loss: 52.7694\n",
      "Epoch 22/32\n",
      "404/404 [==============================] - 0s 339us/sample - loss: 44.2351 - val_loss: 61.1162\n",
      "Epoch 23/32\n",
      "404/404 [==============================] - 0s 310us/sample - loss: 41.8487 - val_loss: 88.2196\n",
      "Epoch 24/32\n",
      "404/404 [==============================] - 0s 346us/sample - loss: 44.3643 - val_loss: 135.8369\n",
      "Epoch 25/32\n",
      "404/404 [==============================] - 0s 321us/sample - loss: 43.7426 - val_loss: 211.8463\n",
      "Epoch 26/32\n",
      "404/404 [==============================] - 0s 269us/sample - loss: 46.4661 - val_loss: 224.9724\n",
      "Epoch 27/32\n",
      "404/404 [==============================] - 0s 213us/sample - loss: 48.3020 - val_loss: 216.6456\n",
      "Epoch 28/32\n",
      "404/404 [==============================] - 0s 206us/sample - loss: 49.9810 - val_loss: 249.0779\n",
      "Epoch 29/32\n",
      "404/404 [==============================] - 0s 216us/sample - loss: 45.7349 - val_loss: 275.3312\n",
      "Epoch 30/32\n",
      "404/404 [==============================] - 0s 242us/sample - loss: 38.8271 - val_loss: 327.9285\n",
      "Epoch 31/32\n",
      "404/404 [==============================] - 0s 224us/sample - loss: 40.5370 - val_loss: 393.2647\n",
      "Epoch 32/32\n",
      "404/404 [==============================] - 0s 205us/sample - loss: 44.4624 - val_loss: 475.7846\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x7f0217d09c50>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train, \n",
    "          batch_size=8, epochs=32, verbose=1, \n",
    "          validation_data=(X_valid, y_valid),\n",
    "          callbacks=[modelcheckpoint]) # val loss below 50 is good, 40 great"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Performing inference"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights(output_path + '/weights.20.hdf5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  9.32909,   0.     ,  18.1    ,   0.     ,   0.713  ,   6.185  ,\n",
       "        98.7    ,   2.2616 ,  24.     , 666.     ,  20.2    , 396.9    ,\n",
       "        18.13   ])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_valid[42]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "14.1"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_valid[42]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[15.265093]], dtype=float32)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.predict(np.reshape(X_valid[42], [1, 13]))"
   ]
  },
  {
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}