Basic Keras Programming Tutorial code.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/shr/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# 0_1. Load keras package needed\n",
    "import numpy as np\n",
    "from keras.utils import np_utils\n",
    "from keras.datasets import mnist\n",
    "from keras import optimizers\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras.layers import Flatten\n",
    "from keras.layers.convolutional import Conv2D\n",
    "from keras.layers.convolutional import MaxPooling2D"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 0_2. Fix random seed\n",
    "np.random.seed(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial train data shape is (60000, 28, 28) Initial train label shape is (60000,)\n"
     ]
    }
   ],
   "source": [
    "# 1_1. Creating Datasets\n",
    "# Load MNIST dataset from server\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "print ( \"Initial train data shape is\",x_train.shape, \"Initial train label shape is\", y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reshape dataset to model input shape \n",
    "x_train = x_train.reshape(60000, 784).astype('float32') / 255.0\n",
    "x_test = x_test.reshape(10000, 784).astype('float32') / 255.0\n",
    "print ( \"Converted train data shape is\",x_train.shape)\n",
    "# Reshape label to one-hat shape for classification problem \n",
    "y_train = np_utils.to_categorical(y_train)\n",
    "y_test = np_utils.to_categorical(y_test)\n",
    "print (\"Converted train label shape is\", y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<svg height=\"221pt\" viewBox=\"0.00 0.00 309.00 221.00\" width=\"309pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g class=\"graph\" id=\"graph0\" transform=\"scale(1 1) rotate(0) translate(4 217)\">\n",
       "<title>G</title>\n",
       "<polygon fill=\"#ffffff\" points=\"-4,4 -4,-217 305,-217 305,4 -4,4\" stroke=\"transparent\"/>\n",
       "<!-- 139685954717288 -->\n",
       "<g class=\"node\" id=\"node1\">\n",
       "<title>139685954717288</title>\n",
       "<polygon fill=\"none\" points=\"0,-166.5 0,-212.5 301,-212.5 301,-166.5 0,-166.5\" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-185.8\">dense_1_input: InputLayer</text>\n",
       "<polyline fill=\"none\" points=\"163,-166.5 163,-212.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"190.5\" y=\"-197.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"163,-189.5 218,-189.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"190.5\" y=\"-174.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"218,-166.5 218,-212.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"259.5\" y=\"-197.3\">(None, 784)</text>\n",
       "<polyline fill=\"none\" points=\"218,-189.5 301,-189.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"259.5\" y=\"-174.3\">(None, 784)</text>\n",
       "</g>\n",
       "<!-- 139685954716728 -->\n",
       "<g class=\"node\" id=\"node2\">\n",
       "<title>139685954716728</title>\n",
       "<polygon fill=\"none\" points=\"30.5,-83.5 30.5,-129.5 270.5,-129.5 270.5,-83.5 30.5,-83.5\" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"81.5\" y=\"-102.8\">dense_1: Dense</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-83.5 132.5,-129.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-114.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"132.5,-106.5 187.5,-106.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"160\" y=\"-91.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-83.5 187.5,-129.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-114.3\">(None, 784)</text>\n",
       "<polyline fill=\"none\" points=\"187.5,-106.5 270.5,-106.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-91.3\">(None, 64)</text>\n",
       "</g>\n",
       "<!-- 139685954717288&#45;&gt;139685954716728 -->\n",
       "<g class=\"edge\" id=\"edge1\">\n",
       "<title>139685954717288-&gt;139685954716728</title>\n",
       "<path d=\"M150.5,-166.3799C150.5,-158.1745 150.5,-148.7679 150.5,-139.8786\" fill=\"none\" stroke=\"#000000\"/>\n",
       "<polygon fill=\"#000000\" points=\"154.0001,-139.784 150.5,-129.784 147.0001,-139.784 154.0001,-139.784\" stroke=\"#000000\"/>\n",
       "</g>\n",
       "<!-- 139685955121952 -->\n",
       "<g class=\"node\" id=\"node3\">\n",
       "<title>139685955121952</title>\n",
       "<polygon fill=\"none\" points=\"34,-.5 34,-46.5 267,-46.5 267,-.5 34,-.5\" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"85\" y=\"-19.8\">dense_2: Dense</text>\n",
       "<polyline fill=\"none\" points=\"136,-.5 136,-46.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"163.5\" y=\"-31.3\">input:</text>\n",
       "<polyline fill=\"none\" points=\"136,-23.5 191,-23.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"163.5\" y=\"-8.3\">output:</text>\n",
       "<polyline fill=\"none\" points=\"191,-.5 191,-46.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-31.3\">(None, 64)</text>\n",
       "<polyline fill=\"none\" points=\"191,-23.5 267,-23.5 \" stroke=\"#000000\"/>\n",
       "<text fill=\"#000000\" font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"229\" y=\"-8.3\">(None, 10)</text>\n",
       "</g>\n",
       "<!-- 139685954716728&#45;&gt;139685955121952 -->\n",
       "<g class=\"edge\" id=\"edge2\">\n",
       "<title>139685954716728-&gt;139685955121952</title>\n",
       "<path d=\"M150.5,-83.3799C150.5,-75.1745 150.5,-65.7679 150.5,-56.8786\" fill=\"none\" stroke=\"#000000\"/>\n",
       "<polygon fill=\"#000000\" points=\"154.0001,-56.784 150.5,-46.784 147.0001,-56.784 154.0001,-56.784\" stroke=\"#000000\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>"
      ],
      "text/plain": [
       "<IPython.core.display.SVG object>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 2_1. Buliding a Model\n",
    "model = Sequential()\n",
    "model.add(Dense(units=64, input_dim=28*28, activation='relu')) # First layer with input shape\n",
    "model.add(Dense(units=10, activation='softmax')) # Second layer with sofrmax activation function\n",
    "\n",
    "# Model Architecture Visualization\n",
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "SVG(model_to_dot(model, show_shapes=True).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 1_2. Creating Datasets\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "print ( \"Initial train data shape is\",x_train.shape, \"Initial train label shape is\", y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reshape dataset to model input shape \n",
    "x_train = x_train.reshape(60000, 28, 28,1).astype('float32') / 255.0\n",
    "x_test = x_test.reshape(10000, 28, 28,1).astype('float32') / 255.0\n",
    "print ( \"Converted train data shape is\",x_train.shape)\n",
    "# Reshape label to one-hat shape for classification problem \n",
    "y_train = np_utils.to_categorical(y_train)\n",
    "y_test = np_utils.to_categorical(y_test)\n",
    "print (\"Converted train label shape is\", y_train.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 2_2. Buliding a Model with Convolutional layer\n",
    "model = Sequential()\n",
    "model.add(Conv2D(32, kernel_size=(3, 3),\n",
    "                 activation='relu',\n",
    "                 input_shape=(28,28,1)))  # First Convolutional layer with input shape \n",
    "model.add(Conv2D(64, (3, 3), activation='relu')) # Second Convolutional layer \n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))  # MaxPooling layer\n",
    "model.add(Flatten())  # Flatten layer for converting CNN codes to Dense layer input \n",
    "model.add(Dense(128, activation='relu')) # hidden Dense layer\n",
    "model.add(Dense(10, activation='softmax')) # Activation layer with sofrmax\n",
    "\n",
    "# Model Architecture Visualization\n",
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "\n",
    "%matplotlib inline\n",
    "\n",
    "SVG(model_to_dot(model, show_shapes=True).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 3. Setting up the Model Learning Process\n",
    "# Define optimizer for training\n",
    "adam = optimizers.Adam(lr=0.0001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)\n",
    "# Compiling sequential model with optimizer\n",
    "model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/shr/anaconda3/lib/python3.6/site-packages/keras/models.py:981: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 48000 samples, validate on 12000 samples\n",
      "Epoch 1/10\n",
      "48000/48000 [==============================] - 2s 35us/step - loss: 0.8283 - acc: 0.7886 - val_loss: 0.3996 - val_acc: 0.8999\n",
      "Epoch 2/10\n",
      "48000/48000 [==============================] - 2s 42us/step - loss: 0.3700 - acc: 0.9017 - val_loss: 0.3092 - val_acc: 0.9174\n",
      "Epoch 3/10\n",
      "48000/48000 [==============================] - 2s 42us/step - loss: 0.3087 - acc: 0.9148 - val_loss: 0.2740 - val_acc: 0.9253\n",
      "Epoch 4/10\n",
      "48000/48000 [==============================] - 2s 42us/step - loss: 0.2770 - acc: 0.9230 - val_loss: 0.2531 - val_acc: 0.9294\n",
      "Epoch 5/10\n",
      "48000/48000 [==============================] - 2s 42us/step - loss: 0.2537 - acc: 0.9291 - val_loss: 0.2348 - val_acc: 0.9356\n",
      "Epoch 6/10\n",
      "48000/48000 [==============================] - 2s 42us/step - loss: 0.2349 - acc: 0.9337 - val_loss: 0.2198 - val_acc: 0.9395\n",
      "Epoch 7/10\n",
      "48000/48000 [==============================] - 3s 53us/step - loss: 0.2194 - acc: 0.9382 - val_loss: 0.2089 - val_acc: 0.9424\n",
      "Epoch 8/10\n",
      "48000/48000 [==============================] - 3s 53us/step - loss: 0.2055 - acc: 0.9421 - val_loss: 0.1989 - val_acc: 0.9444\n",
      "Epoch 9/10\n",
      "48000/48000 [==============================] - 3s 53us/step - loss: 0.1931 - acc: 0.9457 - val_loss: 0.1885 - val_acc: 0.9476\n",
      "Epoch 10/10\n",
      "48000/48000 [==============================] - 3s 53us/step - loss: 0.1824 - acc: 0.9489 - val_loss: 0.1829 - val_acc: 0.9495\n"
     ]
    }
   ],
   "source": [
    "# 4. Training the Model\n",
    "hist = model.fit(x_train, y_train, batch_size=32, validation_split=0.2, shuffle=True, nb_epoch=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0b2bb811d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 5. Confirm the Learning Process\n",
    "# by mathplot model training history can be visualized\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "fig, loss_ax = plt.subplots()\n",
    "acc_ax = loss_ax.twinx()\n",
    "loss_ax.plot(hist.history['loss'], 'y', label='train loss')\n",
    "loss_ax.plot(hist.history['val_loss'], 'r', label='val loss')\n",
    "acc_ax.plot(hist.history['acc'], 'b', label='train acc')\n",
    "acc_ax.plot(hist.history['val_acc'], 'g', label='val acc')\n",
    "loss_ax.set_xlabel('epoch')\n",
    "loss_ax.set_ylabel('loss')\n",
    "acc_ax.set_ylabel('accuray')\n",
    "loss_ax.legend(loc='upper left')\n",
    "acc_ax.legend(loc='lower left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 0s 22us/step\n",
      "evaluation loss and acc\n",
      "[0.1807783282607794, 0.9472]\n"
     ]
    }
   ],
   "source": [
    "# 6. Evaluating the Model\n",
    "loss_and_metrics = model.evaluate(x_test, y_test, batch_size=32)\n",
    "print('evaluation loss and acc')\n",
    "print(loss_and_metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0. 0. 0. 0. 0. 0. 0. 1. 0. 0.]]\n",
      "predicted class\n",
      "[7]\n",
      "class probabilty is\n",
      "[[6.0878265e-05 2.1356923e-08 2.7028471e-04 4.5518465e-03 2.1329497e-07\n",
      "  5.0951239e-05 3.4084209e-09 9.9483889e-01 1.4073811e-05 2.1281923e-04]]\n"
     ]
    }
   ],
   "source": [
    "# 7. Using the Model\n",
    "\n",
    "from keras.models import load_model\n",
    "\n",
    "model.save('mnist_mlp_model.h5')\n",
    "\n",
    "from keras.models import load_model\n",
    "\n",
    "model = load_model('mnist_mlp_model.h5')\n",
    "\n",
    "xhat = x_test[0:1]\n",
    "print(y_test[0:1])\n",
    "yhat = model.predict_classes(xhat)\n",
    "yhatp = model.predict(xhat)\n",
    "print('predicted class')\n",
    "print(yhat)\n",
    "print('class probabilty is')\n",
    "print(yhatp)"
   ]
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}