04_modern_net.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Activation, Dropout\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.datasets import mnist\n",
    "from keras.utils import np_utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "batch_size = 128\n",
    "nb_classes = 10\n",
    "nb_epoch = 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Load MNIST dataset\n",
    "(X_train, y_train), (X_test, y_test) = mnist.load_data()\n",
    "X_train = X_train.reshape(60000, 784)\n",
    "X_test = X_test.reshape(10000, 784)\n",
    "X_train = X_train.astype('float32')\n",
    "X_test = X_test.astype('float32')\n",
    "X_train /= 255\n",
    "X_test /= 255\n",
    "Y_train = np_utils.to_categorical(y_train, nb_classes)\n",
    "Y_test = np_utils.to_categorical(y_test, nb_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "____________________________________________________________________________________________________\n",
      "Layer (type)                     Output Shape          Param #     Connected to                     \n",
      "====================================================================================================\n",
      "dense_1 (Dense)                  (None, 625)           490625      dense_input_1[0][0]              \n",
      "____________________________________________________________________________________________________\n",
      "activation_1 (Activation)        (None, 625)           0           dense_1[0][0]                    \n",
      "____________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)              (None, 625)           0           activation_1[0][0]               \n",
      "____________________________________________________________________________________________________\n",
      "dense_2 (Dense)                  (None, 625)           391250      dropout_1[0][0]                  \n",
      "____________________________________________________________________________________________________\n",
      "activation_2 (Activation)        (None, 625)           0           dense_2[0][0]                    \n",
      "____________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)              (None, 625)           0           activation_2[0][0]               \n",
      "____________________________________________________________________________________________________\n",
      "dense_3 (Dense)                  (None, 10)            6260        dropout_2[0][0]                  \n",
      "____________________________________________________________________________________________________\n",
      "activation_3 (Activation)        (None, 10)            0           dense_3[0][0]                    \n",
      "====================================================================================================\n",
      "Total params: 888135\n",
      "____________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Deep Multilayer Perceptron model\n",
    "model = Sequential()\n",
    "model.add(Dense(output_dim=625, input_dim=784, init='normal'))\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dropout(0.2))\n",
    "model.add(Dense(output_dim=625, input_dim=625, init='normal'))\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(output_dim=10, input_dim=625, init='normal'))\n",
    "model.add(Activation('softmax'))\n",
    "\n",
    "model.compile(optimizer=RMSprop(lr=0.001, rho=0.9), loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "60000/60000 [==============================] - 3s - loss: 0.2623 - acc: 0.9192     \n",
      "Epoch 2/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.1104 - acc: 0.9667     \n",
      "Epoch 3/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0830 - acc: 0.9753     \n",
      "Epoch 4/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0663 - acc: 0.9801     \n",
      "Epoch 5/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0566 - acc: 0.9838     \n",
      "Epoch 6/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0501 - acc: 0.9856     \n",
      "Epoch 7/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0430 - acc: 0.9879     \n",
      "Epoch 8/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0419 - acc: 0.9882     \n",
      "Epoch 9/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0361 - acc: 0.9903     \n",
      "Epoch 10/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0327 - acc: 0.9912     \n",
      "Epoch 11/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0323 - acc: 0.9917     \n",
      "Epoch 12/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0332 - acc: 0.9920     \n",
      "Epoch 13/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0307 - acc: 0.9919     \n",
      "Epoch 14/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0280 - acc: 0.9929     \n",
      "Epoch 15/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0286 - acc: 0.9928     \n",
      "Epoch 16/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0272 - acc: 0.9934     \n",
      "Epoch 17/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0265 - acc: 0.9938     \n",
      "Epoch 18/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0263 - acc: 0.9939     \n",
      "Epoch 19/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0280 - acc: 0.9934     \n",
      "Epoch 20/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0220 - acc: 0.9950     \n",
      "Epoch 21/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0238 - acc: 0.9944     \n",
      "Epoch 22/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0221 - acc: 0.9950     \n",
      "Epoch 23/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0221 - acc: 0.9951     \n",
      "Epoch 24/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0249 - acc: 0.9947     \n",
      "Epoch 25/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0210 - acc: 0.9954     \n",
      "Epoch 26/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0252 - acc: 0.9950     \n",
      "Epoch 27/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0216 - acc: 0.9954     \n",
      "Epoch 28/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0215 - acc: 0.9954     \n",
      "Epoch 29/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0184 - acc: 0.9962     \n",
      "Epoch 30/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0208 - acc: 0.9957     \n",
      "Epoch 31/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0227 - acc: 0.9954     \n",
      "Epoch 32/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0198 - acc: 0.9960     \n",
      "Epoch 33/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0227 - acc: 0.9954     \n",
      "Epoch 34/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0216 - acc: 0.9959     \n",
      "Epoch 35/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0211 - acc: 0.9959     \n",
      "Epoch 36/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0224 - acc: 0.9957     \n",
      "Epoch 37/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0216 - acc: 0.9962     \n",
      "Epoch 38/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0210 - acc: 0.9961     \n",
      "Epoch 39/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0209 - acc: 0.9961     \n",
      "Epoch 40/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0192 - acc: 0.9964     \n",
      "Epoch 41/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0211 - acc: 0.9960     \n",
      "Epoch 42/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0211 - acc: 0.9963     \n",
      "Epoch 43/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0208 - acc: 0.9958     \n",
      "Epoch 44/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0223 - acc: 0.9962     \n",
      "Epoch 45/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0220 - acc: 0.9965     \n",
      "Epoch 46/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0219 - acc: 0.9965     \n",
      "Epoch 47/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0212 - acc: 0.9966     \n",
      "Epoch 48/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0203 - acc: 0.9965     \n",
      "Epoch 49/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0248 - acc: 0.9963     \n",
      "Epoch 50/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0225 - acc: 0.9962     \n",
      "Epoch 51/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0244 - acc: 0.9963     \n",
      "Epoch 52/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0196 - acc: 0.9966     \n",
      "Epoch 53/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0219 - acc: 0.9965     \n",
      "Epoch 54/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0200 - acc: 0.9968     \n",
      "Epoch 55/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0206 - acc: 0.9967     \n",
      "Epoch 56/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0196 - acc: 0.9968     \n",
      "Epoch 57/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0242 - acc: 0.9966     \n",
      "Epoch 58/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0207 - acc: 0.9968     \n",
      "Epoch 59/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0184 - acc: 0.9970     \n",
      "Epoch 60/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0222 - acc: 0.9965     \n",
      "Epoch 61/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0195 - acc: 0.9969     \n",
      "Epoch 62/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0214 - acc: 0.9967     \n",
      "Epoch 63/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0195 - acc: 0.9971     \n",
      "Epoch 64/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0220 - acc: 0.9965     \n",
      "Epoch 65/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0222 - acc: 0.9966     \n",
      "Epoch 66/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0218 - acc: 0.9968     \n",
      "Epoch 67/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0219 - acc: 0.9968     \n",
      "Epoch 68/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0207 - acc: 0.9969     \n",
      "Epoch 69/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0199 - acc: 0.9971     \n",
      "Epoch 70/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0164 - acc: 0.9975     \n",
      "Epoch 71/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0233 - acc: 0.9968     \n",
      "Epoch 72/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0202 - acc: 0.9969     \n",
      "Epoch 73/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0209 - acc: 0.9969     \n",
      "Epoch 74/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0219 - acc: 0.9969     \n",
      "Epoch 75/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0206 - acc: 0.9971     \n",
      "Epoch 76/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0224 - acc: 0.9969     \n",
      "Epoch 77/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0203 - acc: 0.9973     \n",
      "Epoch 78/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0168 - acc: 0.9974     \n",
      "Epoch 79/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0201 - acc: 0.9971     \n",
      "Epoch 80/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0184 - acc: 0.9974     \n",
      "Epoch 81/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0203 - acc: 0.9972     \n",
      "Epoch 82/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0205 - acc: 0.9973     \n",
      "Epoch 83/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0200 - acc: 0.9977     \n",
      "Epoch 84/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0261 - acc: 0.9966     \n",
      "Epoch 85/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0176 - acc: 0.9974     \n",
      "Epoch 86/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0198 - acc: 0.9974     \n",
      "Epoch 87/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0194 - acc: 0.9974     \n",
      "Epoch 88/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0188 - acc: 0.9974     \n",
      "Epoch 89/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0198 - acc: 0.9972     \n",
      "Epoch 90/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0207 - acc: 0.9972     \n",
      "Epoch 91/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0224 - acc: 0.9971     \n",
      "Epoch 92/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0192 - acc: 0.9975     \n",
      "Epoch 93/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0223 - acc: 0.9971     \n",
      "Epoch 94/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0197 - acc: 0.9974     \n",
      "Epoch 95/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0193 - acc: 0.9973     \n",
      "Epoch 96/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0190 - acc: 0.9973     \n",
      "Epoch 97/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0229 - acc: 0.9970     \n",
      "Epoch 98/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0212 - acc: 0.9973     \n",
      "Epoch 99/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0222 - acc: 0.9969     \n",
      "Epoch 100/100\n",
      "60000/60000 [==============================] - 2s - loss: 0.0208 - acc: 0.9972     \n"
     ]
    }
   ],
   "source": [
    "# Train\n",
    "history = model.fit(X_train, Y_train, nb_epoch=nb_epoch, batch_size=batch_size, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 9984/10000 [============================>.] - ETA: 0sSummary: Loss over the test dataset: 0.17, Accuracy: 0.98\n"
     ]
    }
   ],
   "source": [
    "# Evaluate\n",
    "evaluation = model.evaluate(X_test, Y_test, verbose=1)\n",
    "print('Summary: Loss over the test dataset: %.2f, Accuracy: %.2f' % (evaluation[0], evaluation[1]))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3.0
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.4.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}