Recognizer_CNN_keras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# Imports \n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import keras\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten \n",
    "from keras.layers.convolutional import Conv2D\n",
    "from keras.layers import MaxPooling2D\n",
    "from keras.utils import to_categorical\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers.advanced_activations import LeakyReLU\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# read data sets\n",
    "train_data = pd.read_csv('data/train.csv', sep=',', header=0)\n",
    "test_data = pd.read_csv('data/test.csv', sep=',', header=0) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reshape, normalize train and test data for CNN (28 x 28 images)\n",
    "\n",
    "train_label = (train_data.iloc[:, 0]).values\n",
    "train = (train_data.iloc[:, 1:]).values\n",
    "test = test_data\n",
    "# print(np.max(train, axis = 0))\n",
    "\n",
    "n_samples = np.shape(train)[0]\n",
    "n_test = np.shape(test)[0]\n",
    "\n",
    "train = train.reshape(n_samples, 28, 28, 1 )\n",
    "test = test.values.reshape(n_test, 28, 28, 1)\n",
    "\n",
    "# Normalize\n",
    "train = train/ 255\n",
    "test = test/ 255\n",
    "\n",
    "# split the train set into train and validation set\n",
    "train,valid,train_label,valid_label = train_test_split(train, train_label, test_size=0.119047, random_state=13)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(37000, 28, 28, 1)\n",
      "(37000, 10)\n",
      "(5000, 28, 28, 1)\n",
      "(5000, 10)\n"
     ]
    }
   ],
   "source": [
    "#  Check data shape\n",
    "\n",
    "print(np.shape(train))\n",
    "print(np.shape(train_label))\n",
    "print(np.shape(valid))\n",
    "print(np.shape(valid_label))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# initialize training variables\n",
    "\n",
    "batch_size = 64\n",
    "epochs = 20\n",
    "num_classes = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\hihiagar\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\framework\\op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "# Initialize and compile the CNN model (without dropout)\n",
    "\n",
    "model = Sequential()\n",
    "train_label = to_categorical(train_label)\n",
    "\n",
    "model.add(Conv2D(32, kernel_size=(3, 3),activation='linear',input_shape=(28,28,1),padding='same'))\n",
    "model.add(LeakyReLU(alpha=0.1))\n",
    "model.add(MaxPooling2D((2, 2),padding='same'))\n",
    "model.add(Conv2D(32, (3, 3), activation='linear',padding='same'))\n",
    "model.add(LeakyReLU(alpha=0.1))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))\n",
    "model.add(Conv2D(64, (3, 3), activation='linear',padding='same'))\n",
    "model.add(LeakyReLU(alpha=0.1))                  \n",
    "model.add(MaxPooling2D(pool_size=(2, 2),padding='same'))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(128, activation='linear'))\n",
    "model.add(LeakyReLU(alpha=0.1))                  \n",
    "model.add(Dense(num_classes, activation='softmax'))\n",
    "\n",
    "model.compile(loss=keras.losses.categorical_crossentropy, optimizer=keras.optimizers.Adam(),metrics=['accuracy'])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_1 (Conv2D)            (None, 28, 28, 32)        320       \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_1 (LeakyReLU)    (None, 28, 28, 32)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 14, 14, 32)        9248      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_2 (LeakyReLU)    (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 7, 7, 32)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 7, 7, 64)          18496     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_3 (LeakyReLU)    (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 4, 4, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 1024)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 128)               131200    \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_4 (LeakyReLU)    (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                1290      \n",
      "=================================================================\n",
      "Total params: 160,554\n",
      "Trainable params: 160,554\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# check mmodel parameters at each layer\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\hihiagar\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\ops\\math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "Train on 37000 samples, validate on 5000 samples\n",
      "Epoch 1/20\n",
      "37000/37000 [==============================] - 70s 2ms/step - loss: 0.2340 - acc: 0.9274 - val_loss: 0.0649 - val_acc: 0.9792\n",
      "Epoch 2/20\n",
      "37000/37000 [==============================] - 70s 2ms/step - loss: 0.0559 - acc: 0.9818 - val_loss: 0.0559 - val_acc: 0.9828\n",
      "Epoch 3/20\n",
      "37000/37000 [==============================] - 65s 2ms/step - loss: 0.0395 - acc: 0.9877 - val_loss: 0.0491 - val_acc: 0.9862\n",
      "Epoch 4/20\n",
      "37000/37000 [==============================] - 65s 2ms/step - loss: 0.0297 - acc: 0.9911 - val_loss: 0.0476 - val_acc: 0.9872\n",
      "Epoch 5/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0219 - acc: 0.9926 - val_loss: 0.0735 - val_acc: 0.9792\n",
      "Epoch 6/20\n",
      "37000/37000 [==============================] - 65s 2ms/step - loss: 0.0177 - acc: 0.9941 - val_loss: 0.0517 - val_acc: 0.9876\n",
      "Epoch 7/20\n",
      "37000/37000 [==============================] - 70s 2ms/step - loss: 0.0167 - acc: 0.9942 - val_loss: 0.0487 - val_acc: 0.9892\n",
      "Epoch 8/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0131 - acc: 0.9959 - val_loss: 0.0494 - val_acc: 0.9870\n",
      "Epoch 9/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0108 - acc: 0.9961 - val_loss: 0.0505 - val_acc: 0.9878\n",
      "Epoch 10/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0111 - acc: 0.9967 - val_loss: 0.0590 - val_acc: 0.9866\n",
      "Epoch 11/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0105 - acc: 0.9962 - val_loss: 0.0641 - val_acc: 0.9842\n",
      "Epoch 12/20\n",
      "37000/37000 [==============================] - 69s 2ms/step - loss: 0.0078 - acc: 0.9974 - val_loss: 0.0659 - val_acc: 0.9864\n",
      "Epoch 13/20\n",
      "37000/37000 [==============================] - 76s 2ms/step - loss: 0.0085 - acc: 0.9972 - val_loss: 0.0606 - val_acc: 0.9858\n",
      "Epoch 14/20\n",
      "37000/37000 [==============================] - 78s 2ms/step - loss: 0.0073 - acc: 0.9977 - val_loss: 0.0692 - val_acc: 0.9886\n",
      "Epoch 15/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0074 - acc: 0.9980 - val_loss: 0.0658 - val_acc: 0.9870\n",
      "Epoch 16/20\n",
      "37000/37000 [==============================] - 63s 2ms/step - loss: 0.0046 - acc: 0.9985 - val_loss: 0.0571 - val_acc: 0.9884\n",
      "Epoch 17/20\n",
      "37000/37000 [==============================] - 59s 2ms/step - loss: 0.0049 - acc: 0.9983 - val_loss: 0.0584 - val_acc: 0.9896\n",
      "Epoch 18/20\n",
      "37000/37000 [==============================] - 61s 2ms/step - loss: 0.0089 - acc: 0.9969 - val_loss: 0.0624 - val_acc: 0.9878\n",
      "Epoch 19/20\n",
      "37000/37000 [==============================] - 67s 2ms/step - loss: 0.0041 - acc: 0.9988 - val_loss: 0.0688 - val_acc: 0.9878\n",
      "Epoch 20/20\n",
      "37000/37000 [==============================] - 72s 2ms/step - loss: 0.0044 - acc: 0.9987 - val_loss: 0.0667 - val_acc: 0.9900\n"
     ]
    }
   ],
   "source": [
    "# One-hot for validation labels\n",
    "valid_label = to_categorical(valid_label)\n",
    "\n",
    "# Train the model\n",
    "lossF = model.fit(train, train_label, batch_size=batch_size,epochs=epochs,verbose=1,validation_data=(valid, valid_label))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "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"
    },
    {
     "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": [
    "# Plot training,validation accuracy & loss functions\n",
    "\n",
    "accuracy = lossF.history['acc']\n",
    "val_accuracy = lossF.history['val_acc']\n",
    "loss = lossF.history['loss']\n",
    "val_loss = lossF.history['val_loss']\n",
    "epochs = range(len(accuracy))\n",
    "plt.plot(epochs, accuracy, 'bo', label='Training accuracy')\n",
    "plt.plot(epochs, val_accuracy, 'b', label='Validation accuracy')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.legend()\n",
    "plt.figure()\n",
    "plt.plot(epochs, loss, 'bo', label='Training loss')\n",
    "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "# predict the test classes\n",
    "predCl = model.predict_classes(test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "#  Write the result to a CSV submission file\n",
    "\n",
    "j=1\n",
    "with open(\"data/result.csv\", \"w+\") as f:\n",
    "    f.write('ImageId,Label')\n",
    "    f.write('\\n')\n",
    "    for i in predCl:\n",
    "        f.write(''+ (str)(j) + ',' + (str)(i))\n",
    "        j+=1\n",
    "        f.write('\\n')"
   ]
  }
 ],
 "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}