UG_ASSIGN_5.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "bhZts28k-Mfz"
   },
   "source": [
    "# Convolutional Model with Batch Normalization\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 34
    },
    "colab_type": "code",
    "id": "-bMSmAb2-YLz",
    "outputId": "6e0ce248-87bd-4782-e263-58b83126148c"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "from keras.models import Sequential\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers.convolutional import Conv2D\n",
    "from keras.layers.convolutional import MaxPooling2D\n",
    "from keras.layers.core import Activation\n",
    "from keras.layers.core import Flatten\n",
    "from keras.layers.core import Dropout\n",
    "from keras.layers.core import Dense\n",
    "from keras import backend as K\n",
    "from sklearn.metrics import classification_report\n",
    "from keras.optimizers import SGD\n",
    "from keras.datasets import fashion_mnist\n",
    "from keras.utils import np_utils\n",
    "from keras import backend as K\n",
    "from imutils import build_montages\n",
    "import matplotlib.pyplot as plt\n",
    "import cv2\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1244
    },
    "colab_type": "code",
    "id": "V2YzCIE--pQe",
    "outputId": "398b2cad-f84d-4007-e342-84b7a9369dc4"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING: Logging before flag parsing goes to stderr.\n",
      "W0619 06:41:22.278541 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:74: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
      "\n",
      "W0619 06:41:22.326915 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:517: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
      "\n",
      "W0619 06:41:22.335511 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4138: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
      "\n",
      "W0619 06:41:22.374803 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:174: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.\n",
      "\n",
      "W0619 06:41:22.375838 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:181: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.\n",
      "\n",
      "W0619 06:41:25.125357 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1834: The name tf.nn.fused_batch_norm is deprecated. Please use tf.compat.v1.nn.fused_batch_norm instead.\n",
      "\n",
      "W0619 06:41:25.378306 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3976: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.\n",
      "\n",
      "W0619 06:41:25.387835 139707850680192 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_1 (Conv2D)            (None, 28, 28, 32)        320       \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    (None, 28, 28, 32)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 28, 28, 32)        9248      \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 28, 28, 32)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 28, 28, 32)        128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 14, 14, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 14, 14, 64)        18496     \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 14, 14, 64)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 14, 14, 64)        36928     \n",
      "_________________________________________________________________\n",
      "activation_4 (Activation)    (None, 14, 14, 64)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_4 (Batch (None, 14, 14, 64)        256       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 7, 7, 64)          0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 3136)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 512)               1606144   \n",
      "_________________________________________________________________\n",
      "activation_5 (Activation)    (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                5130      \n",
      "_________________________________________________________________\n",
      "activation_6 (Activation)    (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 1,679,082\n",
      "Trainable params: 1,677,674\n",
      "Non-trainable params: 1,408\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#width=28, height=28, depth=1, classes=10\n",
    "model = Sequential()\n",
    "inputShape = (28, 28, 1)\n",
    "chanDim = -1\n",
    "# first CONV => RELU => CONV => RELU => POOL layer set\n",
    "model.add(Conv2D(32, (3, 3), padding=\"same\",\n",
    "  input_shape=inputShape))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization(axis=chanDim))\n",
    "model.add(Conv2D(32, (3, 3), padding=\"same\"))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization(axis=chanDim))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "# second CONV => RELU => CONV => RELU => POOL layer set\n",
    "model.add(Conv2D(64, (3, 3), padding=\"same\"))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization(axis=chanDim))\n",
    "model.add(Conv2D(64, (3, 3), padding=\"same\"))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization(axis=chanDim))\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "# first (and only) set of FC => RELU layers\n",
    "model.add(Flatten())\n",
    "model.add(Dense(512))\n",
    "model.add(Activation(\"relu\"))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Dropout(0.5))\n",
    "\n",
    "# softmax classifier\n",
    "model.add(Dense(10))\n",
    "model.add(Activation(\"softmax\"))\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "JbgeAjRJCuWS"
   },
   "source": [
    "##Downloading Image Input and the remaining procesing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 170
    },
    "colab_type": "code",
    "id": "0JgOE9I0KPox",
    "outputId": "021db0d2-64e9-41ee-c5ff-efa5a8574079"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz\n",
      "32768/29515 [=================================] - 0s 3us/step\n",
      "Downloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz\n",
      "26427392/26421880 [==============================] - 2s 0us/step\n",
      "Downloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz\n",
      "8192/5148 [===============================================] - 0s 0us/step\n",
      "Downloading data from http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz\n",
      "4423680/4422102 [==============================] - 1s 0us/step\n",
      "(60000, 28, 28)\n"
     ]
    }
   ],
   "source": [
    "NUM_EPOCHS = 25      # Number of epochs\n",
    "INIT_LR = 1e-2      # Initial Learning rate\n",
    "BS = 32    # Batch size\n",
    "((X_train, y_train), (X_test, y_test)) = fashion_mnist.load_data()\n",
    "print(X_train.shape)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "TCynjzlRKTFc"
   },
   "source": [
    "### Data Augmentation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "ZzRoAWChKQNC"
   },
   "outputs": [],
   "source": [
    "y_train = y_train.reshape((y_train.shape[0], 1))\n",
    "#print(train_labels.shape)\n",
    "rows = 28\n",
    "cols = 28\n",
    "\n",
    "def translation(train_images):\n",
    "    M = np.float32([[1,0,5],[0,1,10]])\n",
    "    dst = cv2.warpAffine(train_images,M,(cols,rows))\n",
    "    dst = dst.reshape((1, dst.shape[0], dst.shape[1]))\n",
    "    #print(dst.shape)\n",
    "    return dst\n",
    "\n",
    "\n",
    "for i in range (500):\n",
    "    X_train = np.vstack((X_train, translation(X_train[i])))\n",
    "    y_train = np.append(y_train, y_train[i])\n",
    "\n",
    "y_train = y_train.reshape((y_train.shape[0], 1))   \n",
    "\n",
    "def rotation(train_images, angle):\n",
    "    M = cv2.getRotationMatrix2D((cols/2,rows/2),-75,1)\n",
    "    rot = cv2.warpAffine(train_images[i],M,(cols,rows))\n",
    "    return rot\n",
    "\n",
    "for i in range (500):\n",
    "    X_train = np.vstack((X_train, translation(X_train[i])))\n",
    "    y_train = np.append(y_train, y_train[i])\n",
    "\n",
    "y_train = y_train.reshape((y_train.shape[0], 1)) \n",
    "def rotation(train_images, angle):\n",
    "    M = cv2.getRotationMatrix2D((cols/2,rows/2),180,1)\n",
    "    rot = cv2.warpAffine(train_images[i],M,(cols,rows))\n",
    "    return rot\n",
    "\n",
    "for i in range (500):\n",
    "    X_train = np.vstack((X_train, translation(X_train[i])))\n",
    "    y_train = np.append(y_train, y_train[i])\n",
    "\n",
    "y_train = y_train.reshape((y_train.shape[0], 1)) \n",
    "\n",
    "def rotation(train_images, angle):\n",
    "    M = cv2.getRotationMatrix2D((cols/2,rows/2),75,1)\n",
    "    rot = cv2.warpAffine(train_images[i],M,(cols,rows))\n",
    "    return rot\n",
    "\n",
    "for i in range (500):\n",
    "    X_train = np.vstack((X_train, translation(X_train[i])))\n",
    "    y_train = np.append(y_train, y_train[i])\n",
    "\n",
    "y_train = y_train.reshape((y_train.shape[0], 1)) \n",
    "\n",
    "\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "MSrM5Qts_Bl9"
   },
   "outputs": [],
   "source": [
    "# if we are using \"channels first\" ordering, then reshape the design\n",
    "# matrix such that the matrix is:\n",
    "# \tnum_samples x depth x rows x columns\n",
    "if K.image_data_format() == \"channels_first\":\n",
    "\tX_train = X_train.reshape((X_train.shape[0], 1, 28, 28))\n",
    "\tX_test = X_test.reshape((X_test.shape[0], 1, 28, 28))\n",
    "else:\n",
    "\tX_train = X_train.reshape((X_train.shape[0], 28, 28, 1))\n",
    "\tX_test = X_test.reshape((X_test.shape[0], 28, 28, 1))\n",
    "  \n",
    "X_train = X_train.astype(\"float32\") / 255.0\n",
    "X_test = X_test.astype(\"float32\") / 255.0\n",
    "# one-hot encode the training and testing labels\n",
    "y_train = np_utils.to_categorical(y_train, 10)\n",
    "y_test = np_utils.to_categorical(y_test, 10)\n",
    "# initialize the label names\n",
    "labelNames = [\"top\", \"trouser\", \"pullover\", \"dress\", \"coat\",\n",
    "\t\"sandal\", \"shirt\", \"sneaker\", \"bag\", \"ankle boot\"]\n",
    "# initialize the optimizer and model\n",
    "opt = SGD(lr=INIT_LR, momentum=0.9, decay=INIT_LR / NUM_EPOCHS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 71
    },
    "colab_type": "code",
    "id": "7BIG5kbQDGjy",
    "outputId": "2b0e9c37-375b-48e7-9b26-058e0d93b714"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "W0619 06:41:59.805603 139707850680192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/optimizers.py:790: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "model.compile(loss=\"categorical_crossentropy\", optimizer=opt,\n",
    "\tmetrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 955
    },
    "colab_type": "code",
    "id": "9dT-O0bYDXh-",
    "outputId": "bce15084-c0c0-4472-8c2f-da83be5fb337"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "W0619 06:41:59.922530 139707850680192 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 62000 samples, validate on 10000 samples\n",
      "Epoch 1/25\n",
      "62000/62000 [==============================] - 22s 357us/step - loss: 0.5763 - acc: 0.8078 - val_loss: 0.3497 - val_acc: 0.8765\n",
      "Epoch 2/25\n",
      "62000/62000 [==============================] - 18s 283us/step - loss: 0.3562 - acc: 0.8731 - val_loss: 0.3465 - val_acc: 0.8676\n",
      "Epoch 3/25\n",
      "62000/62000 [==============================] - 18s 284us/step - loss: 0.3029 - acc: 0.8919 - val_loss: 0.2525 - val_acc: 0.9108\n",
      "Epoch 4/25\n",
      "62000/62000 [==============================] - 18s 290us/step - loss: 0.2779 - acc: 0.9006 - val_loss: 0.2422 - val_acc: 0.9117\n",
      "Epoch 5/25\n",
      "62000/62000 [==============================] - 17s 281us/step - loss: 0.2572 - acc: 0.9065 - val_loss: 0.2291 - val_acc: 0.9179\n",
      "Epoch 6/25\n",
      "62000/62000 [==============================] - 17s 281us/step - loss: 0.2459 - acc: 0.9108 - val_loss: 0.2338 - val_acc: 0.9146\n",
      "Epoch 7/25\n",
      "62000/62000 [==============================] - 17s 281us/step - loss: 0.2383 - acc: 0.9132 - val_loss: 0.2172 - val_acc: 0.9223\n",
      "Epoch 8/25\n",
      "62000/62000 [==============================] - 18s 286us/step - loss: 0.2304 - acc: 0.9158 - val_loss: 0.2109 - val_acc: 0.9241\n",
      "Epoch 9/25\n",
      "62000/62000 [==============================] - 17s 280us/step - loss: 0.2206 - acc: 0.9197 - val_loss: 0.2107 - val_acc: 0.9226\n",
      "Epoch 10/25\n",
      "62000/62000 [==============================] - 17s 280us/step - loss: 0.2159 - acc: 0.9224 - val_loss: 0.2050 - val_acc: 0.9268\n",
      "Epoch 11/25\n",
      "62000/62000 [==============================] - 17s 280us/step - loss: 0.2102 - acc: 0.9235 - val_loss: 0.2055 - val_acc: 0.9236\n",
      "Epoch 12/25\n",
      "62000/62000 [==============================] - 17s 280us/step - loss: 0.2049 - acc: 0.9244 - val_loss: 0.2066 - val_acc: 0.9247\n",
      "Epoch 13/25\n",
      "62000/62000 [==============================] - 18s 285us/step - loss: 0.2021 - acc: 0.9260 - val_loss: 0.2137 - val_acc: 0.9224\n",
      "Epoch 14/25\n",
      "62000/62000 [==============================] - 18s 284us/step - loss: 0.1993 - acc: 0.9276 - val_loss: 0.1979 - val_acc: 0.9281\n",
      "Epoch 15/25\n",
      "62000/62000 [==============================] - 18s 284us/step - loss: 0.1951 - acc: 0.9291 - val_loss: 0.1944 - val_acc: 0.9301\n",
      "Epoch 16/25\n",
      "62000/62000 [==============================] - 18s 291us/step - loss: 0.1912 - acc: 0.9297 - val_loss: 0.1933 - val_acc: 0.9297\n",
      "Epoch 17/25\n",
      "62000/62000 [==============================] - 18s 286us/step - loss: 0.1897 - acc: 0.9315 - val_loss: 0.1921 - val_acc: 0.9303\n",
      "Epoch 18/25\n",
      "62000/62000 [==============================] - 18s 284us/step - loss: 0.1866 - acc: 0.9318 - val_loss: 0.1946 - val_acc: 0.9279\n",
      "Epoch 19/25\n",
      "62000/62000 [==============================] - 17s 282us/step - loss: 0.1865 - acc: 0.9324 - val_loss: 0.1914 - val_acc: 0.9313\n",
      "Epoch 20/25\n",
      "62000/62000 [==============================] - 17s 281us/step - loss: 0.1821 - acc: 0.9335 - val_loss: 0.1895 - val_acc: 0.9311\n",
      "Epoch 21/25\n",
      "62000/62000 [==============================] - 17s 282us/step - loss: 0.1808 - acc: 0.9337 - val_loss: 0.1898 - val_acc: 0.9311\n",
      "Epoch 22/25\n",
      "62000/62000 [==============================] - 18s 288us/step - loss: 0.1775 - acc: 0.9355 - val_loss: 0.1878 - val_acc: 0.9319\n",
      "Epoch 23/25\n",
      "62000/62000 [==============================] - 17s 281us/step - loss: 0.1781 - acc: 0.9345 - val_loss: 0.1883 - val_acc: 0.9327\n",
      "Epoch 24/25\n",
      "62000/62000 [==============================] - 17s 280us/step - loss: 0.1748 - acc: 0.9364 - val_loss: 0.1850 - val_acc: 0.9316\n",
      "Epoch 25/25\n",
      "62000/62000 [==============================] - 18s 283us/step - loss: 0.1745 - acc: 0.9365 - val_loss: 0.1862 - val_acc: 0.9336\n"
     ]
    }
   ],
   "source": [
    "H = model.fit(X_train, y_train,\n",
    "\tvalidation_data=(X_test, y_test),\n",
    "\tbatch_size=BS, epochs=NUM_EPOCHS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 588
    },
    "colab_type": "code",
    "id": "zD9Kk5ekDfBO",
    "outputId": "649f8158-61da-4915-ded8-002591d10652"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "         top       0.90      0.86      0.88      1000\n",
      "     trouser       1.00      0.98      0.99      1000\n",
      "    pullover       0.93      0.89      0.91      1000\n",
      "       dress       0.93      0.94      0.94      1000\n",
      "        coat       0.88      0.92      0.90      1000\n",
      "      sandal       0.99      0.98      0.99      1000\n",
      "       shirt       0.78      0.81      0.80      1000\n",
      "     sneaker       0.96      0.98      0.97      1000\n",
      "         bag       0.99      0.99      0.99      1000\n",
      "  ankle boot       0.98      0.97      0.98      1000\n",
      "\n",
      "    accuracy                           0.93     10000\n",
      "   macro avg       0.93      0.93      0.93     10000\n",
      "weighted avg       0.93      0.93      0.93     10000\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEaCAYAAAAL7cBuAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRFESIIQQQhT1f+VfmcClwMVbAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# make predictions on the test set\n",
    "preds = model.predict(X_test)\n",
    "\n",
    "# show a nicely formatted classification report\n",
    "print(classification_report(y_test.argmax(axis=1), preds.argmax(axis=1),\n",
    "\ttarget_names=labelNames))\n",
    "\n",
    "# plot the training loss and accuracy\n",
    "N = NUM_EPOCHS\n",
    "plt.style.use(\"ggplot\")\n",
    "plt.figure()\n",
    "plt.plot(np.arange(0, N), H.history[\"loss\"], label=\"train_loss\")\n",
    "plt.plot(np.arange(0, N), H.history[\"val_loss\"], label=\"val_loss\")\n",
    "plt.plot(np.arange(0, N), H.history[\"acc\"], label=\"train_acc\")\n",
    "plt.plot(np.arange(0, N), H.history[\"val_acc\"]\n",
    "         , label=\"val_acc\")\n",
    "plt.title(\"Training Loss and Accuracy on Dataset\")\n",
    "plt.xlabel(\"Epoch #\")\n",
    "plt.ylabel(\"Loss/Accuracy\")\n",
    "plt.legend(loc=\"lower left\")\n",
    "plt.savefig(\"plot.png\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "lbOCvbZXqcBc"
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "u1kqS5My22jg"
   },
   "source": [
    "##Note the Following results for number of epochs == 25\n",
    "# Some of the inferences:\n",
    "\n",
    "Without Augmentation :  Train_Acc : 93.8 Test_Acc : 93.4\n",
    "\n",
    "\n",
    "With Data Augmentation: Train_Acc : 89.15 Test_Acc :82.60\n",
    "\n",
    "\n",
    "Without Activation (relu ) : Train_Acc : 91.37 Test_Acc = 91.11\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "_AAcdjXk3phX"
   },
   "source": [
    "## SUMMARY\n",
    "\n",
    "It is clearly depicted that introducing augmentation only to the training set introduce lot of variance and made the distribution of the train data a bit different than the test set . And note the accuracy dropped by 4% approx. \n",
    "And the model without activations are also tested and got a accuracy reduction of 2% approx. From this we can infere that introduction of further non linearity improves the model performance atleaast in this case.\n",
    "\n",
    "Dropouts in this model plays a important role in improving the accuracy.\n"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "name": "Assign_5.ipynb",
   "provenance": [],
   "toc_visible": true,
   "version": "0.3.2"
  },
  "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}