Image_classification_in_SVHN_Dataset.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "YZvtavUvSxBp"
   },
   "source": [
    "# 1. Import the Libraries "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "LfvotWHRUZyP"
   },
   "outputs": [],
   "source": [
    "#Importing important libraries\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import h5py\n",
    "import matplotlib.pyplot as plt\n",
    "#To enable plotting graphs in Jupyter notebook\n",
    "%matplotlib inline\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.metrics import confusion_matrix,accuracy_score\n",
    "from sklearn.metrics import classification_report\n",
    "from keras.models import Sequential\n",
    "from keras.utils.np_utils import to_categorical\n",
    "from keras.layers import Activation, Dense\n",
    "from keras.layers import BatchNormalization, Dropout\n",
    "from keras import optimizers\n",
    "import tensorflow as tf\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "8QlH23LWUZyR"
   },
   "source": [
    "# 2. Load the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "WR_XhXkuqaCv"
   },
   "outputs": [],
   "source": [
    "#Open the file as readonly\n",
    "data1=h5py.File(\"D:\\\\D\\\\1. Great Lakes 18-19\\\\4. Deep Learning and NN\\\\Week4\\\\SVHN_single_grey1.h5\",'r')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 35
    },
    "colab_type": "code",
    "id": "Yf46WImxUZyV",
    "outputId": "4fc19ec4-da6e-4382-f732-05f5db7aadd1"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<KeysViewHDF5 ['X_test', 'X_train', 'X_val', 'y_test', 'y_train', 'y_val']>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data1.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "sNnu7PxaUZyX"
   },
   "outputs": [],
   "source": [
    "#Load the training, testing, and validation data\n",
    "X_train=data1['X_train']\n",
    "X_test=data1['X_test']\n",
    "X_val=data1['X_val']\n",
    "y_train=data1['y_train']\n",
    "y_test=data1['y_test']\n",
    "y_val=data1['y_val']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 121
    },
    "colab_type": "code",
    "id": "Ta-rmIzCUZyZ",
    "outputId": "a74697fa-99be-4a4f-96cb-856857cf0215"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training data X-- Shape : (42000, 32, 32) and Data Type :  float32\n",
      "Testing data X-- Shape : (18000, 32, 32) and Data Type :  float32\n",
      "Validation data X-- Shape : (60000, 32, 32) and Data Type :  float32\n",
      "Training data y-- Shape : (42000,) and Data Type :  uint8\n",
      "Testing data y-- Shape : (18000,) and Data Type :  uint8\n",
      "Validation data y-- Shape : (60000,) and Data Type :  uint8\n"
     ]
    }
   ],
   "source": [
    "# Printing the shape and data type of training, testing, and validation data\n",
    "print(\"Training data X-- Shape :\", X_train.shape,\"and Data Type : \", X_train.dtype)\n",
    "print(\"Testing data X-- Shape :\", X_test.shape,\"and Data Type : \", X_test.dtype)\n",
    "print(\"Validation data X-- Shape :\", X_val.shape,\"and Data Type : \", X_val.dtype)\n",
    "print(\"Training data y-- Shape :\", y_train.shape,\"and Data Type : \", y_train.dtype)\n",
    "print(\"Testing data y-- Shape :\", y_test.shape,\"and Data Type : \", y_test.dtype)\n",
    "print(\"Validation data y-- Shape :\", y_val.shape,\"and Data Type : \", y_val.dtype)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Visualizing the dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 486
    },
    "colab_type": "code",
    "id": "PCepVJhmUZyd",
    "outputId": "c738672b-9a73-4c8a-d80f-b87babc497a5"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 100 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig=plt.figure(figsize=(8,8))\n",
    "columns=10\n",
    "rows=10\n",
    "for i in range(1, columns*rows+1):\n",
    "    img=X_test[i]\n",
    "    fig.add_subplot(rows,columns,i)\n",
    "    plt.imshow(img,cmap='gray')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "QToVz5t6UZzC",
    "outputId": "f80e0e45-9c89-47e5-98e5-c6ae27e24790"
   },
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Label:  2\n"
     ]
    }
   ],
   "source": [
    "# show the number in the dataset\n",
    "plt.imshow(X_train[0],cmap='gray')    \n",
    "plt.show()\n",
    "print('Label: ', y_train[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "NXb9mX72UZzF",
    "outputId": "708b54d2-bbfe-4391-fa8b-afb9215c0e5e"
   },
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Label:  1\n"
     ]
    }
   ],
   "source": [
    "# show the number in the dataset\n",
    "plt.imshow(X_test[0],cmap='gray')    \n",
    "plt.show()\n",
    "print('Label: ', y_test[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "5-8Z7z55UZzJ",
    "outputId": "0684a65a-39ed-4d82-9e0f-ab02b0a32b0c"
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAjwAAAA9CAYAAACpzLMWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nO19WY9c13Xuqnmeiz2ym02ySYoiRVOURFmKHMpxpMiCE8d2DARQXhIDmZCXPOYtec0/MBAEyYuTOEjiAE4kJZBki5pMmoMoUZQ4dJPsbpLd1V1d8zzch8L6+O2j011Vuffi4hJnvahUrD5nj2uv/X1rcPX7fXHEEUccccQRRxx5lMX9/7oBjjjiiCOOOOKII/+3xTF4HHHEEUccccSRR14cg8cRRxxxxBFHHHnkxTF4HHHEEUccccSRR14cg8cRRxxxxBFHHHnkxTF4HHHEEUccccSRR168u/3j1NRU3+VyiYhIOByWqakpERGZm5uTeDwuIiKtVks2NjZERGR5eVm2trbw95lMRkREpqen5dChQyIicuzYMYnFYiIiUiwW5e7duyIicu3aNfn8889FRKRQKIjbPbDFfD6feDwevPerX/2qiIg899xz8thjj6FtGl6/sbEhn332mYiI/OEf/qFr2AC8/fbb6KPb7cZ7PR6P9Ho9fFZpNpuiv3e5XBiTAwcOiNc7GM5er4f28O+73S6e2e/3jefqewOBAD53Oh1pt9v42y+++EJERCqVCt71wgsvDO3jD3/4w34wGBQRkXw+LysrKyIyGGd9vtfrxTNbrZa0Wi20X/vicrnE5/PhudqXer0utVoN7ex2u19qQ7fblU6ng76GQiEREUkkElgn2WxWstmsiAzmVH/zgx/8YNc+/t3f/V3/wYMHIiISDAYlHA6LiEgymcTnYDBotJ3nR9fsxsaGbG9vi4hIu92WaDSKdmlbXC6X6FhGIhHx+/14po5lq9XC83ksXC4X/r/T6WD+v//97w+dwx//+Md9fWav18N493o9CQQCIiLi9/uNNdhsNtEunatOp2Osa1772v5qtSr1eh190edsbm7KjRs3MFb6m06ng3GYnJyU+fl5ERHZs2cPxvCv//qvh/bx7t27/Xv37omIyD/8wz/IW2+9JSIi29vbUi6XRUTk2WeflT/6oz8SEZFTp05hPEOhkCSTSTxL+9jr9dDHTqeDeWm1Wvg+FAoJj60+s9/vy61bt0RE5I033pCPPvpIRES2trYkn8+LyGDt6zgUi8WhffyLv/gLzGOn05FGoyEigz1dqVRERKRcLuNzrVbDb+r1ujGn+hy/34917vP50K9Go4F10ul0oFf498FgEHNn3bu6Pj0eD/72/Pnzu/ZxY2Ojr/us3+9jz+t7RQZjrM9uNpvQNcFgEO/pdru2n6vVquzZswfPWVtbw7t4T6hOiUajUq1WMWb6fSgUkkKhICKDta9zuLCwMHQOz5w509e1lslkoFdYR9+7d082NzfxN3NzcyIyOP8ef/xxERGZmZmBvnG73di7vC/7/b5xZug+2NjYkDt37oiIyPr6OtZIOByGDs1kMoYe0v7+9Kc/HdrH3/u93+uvrq5i3HS9NxoNzAWfbYlEAu3nOa3Vamh/KBTCWHk8HuF1otJut/G3vH4CgQCeMzs7K9/97ndFROTll1+GLcJtW1xctO2jg/A44ogjjjjiiCOPvOyK8LjdbliOCwsLcubMGRERefLJJ2Fl1+t1OXfunIgMEBtFe7LZrDz11FMiIvL888/Lk08+KSKDW59a+m63Gxbr2bNn5e233xYRkcuXL+M5nU4HN7eTJ0/Kb/7mb4qIyJEjR9DOQqGAW+7Ro0dldnZ25AFgNIYRDJfLZViejOroLajZbGJ8qtUqbv6BQMB4plqsbNXq/4sMbs6KdHU6HeOmp7eE+/fvG9+rtf7CCy8M7WMoFMINJpfLid6iV1dXcWNkYcu62WwaKJNa0PpfEfOWG4/HDdSDEQd9jsvlwvf1eh1j5fP5gKSEw2FjrHYTRs74Rs9z63a78dnj8RjvVMSRkYput4u2pFIpPLNSqWBsqtUqPnNb/X4/nl+v1zHPvV7PQBAZ4RsmbrfbQGl0LPm2HAgEsA9cLpdxS9R2ulwuPIfH2+PxYA7r9bqBROnfcvv1Gfq32n+3243f841rFOl2u2gzz1E4HEZ7GHnltgWDQXz2er3GftU9WqvV8Jx6vY41WKvVMO9+vx9j2Gq1JJfLiYjI559/DgS63++jX+1223YP7SRerxdt4Dni+W2328b4282Lx+NBOyORiEQiERExb8KNRsPoo+qher2Oz4lEAn3Xtmgf7do5imh7/X6/oS+0LY1Gw0CP+DOjHPp9p9PB+8PhMPRgPp+X+/fv4/mMtOj3PDa61vU3+v+8FkaRWCyGM2ZmZgbPabVa2JciYqyLVColIiITExOyd+9eERHZv38/0NB+v28gkbpm2+22oX8V0fL5fDgD8vk8ft/pdLAukskk0J5ms4kxGUVeeeUVoIArKytAN2/evInx57OwVCqhncFgEKhLOp2Gfg0EAvh9uVwGulUoFHB2BoNBzBezAoz8VKtVo78qvGZ3kl1nOR6PY8FOT0+DQtq/f78kEgkRGRyg+tJWq4WFs3//fhzGzz77LBbFrVu3AD0eOHBAJicnRUTkm9/8Jga40WhgMkVEvvKVr4iIyEsvvQTDqdfryYcffigiIu+88w4m9uWXX5aDBw/u2mkWt9uNjcWQLivxVqsFw+PGjRuALbe2trC4Dhw4gLY988wzsn//fhEZbDIdQz54+v2+lEolERH5+7//e/nXf/1XtIkPDJ3YRqNhGFHjKKBkMonF0uv18N5isWhA3mwc8AGuf8uLy+v1GoeZ/j4YDGINWGky/Xs+qLxeLzYKK/p+v28YnLsJ/x0rHTZ+RB4amDx+bJz0+33Mp/ZFZACLqxJ3uVwGNccKTsfA6/UaY6lipTGHbU6r2MHc3W4X7/J6vWiz2+3G8xuNBj5ze/1+PygnnqtarWaMlfbB5XLhezYg3W43Dk0+OJgeGkV8Ph90QDqdxjMrlQr0zfz8PH7TarWgTF0uF+YoFAoZh7uu9wcPHsj6+rqIDC4Z2rZoNAo9tLi4KAsLCxgrhfJLpRIOsGazibb1+33jIB0m1vHcSbRtvV7PoJf1vaFQCGMSj8fxORQK4bmtVgt9L5VKuFxWKhU8kw8VviD8T4XXpogYFycdJ6/XaxhZvF+ZvmF9wS4OemFbW1vD30YiEUmn0/i90tTFYhGHbyKRwDNrtRrW6p49e3CIjyLJZBKGysGDB7Hn2CCt1Wo4M9jw570SiUSM/cqi+5Lng/dToVCAYeDz+aB7EokEjLEDBw7IgQMH8HvWbaOIghrpdBpryuv1yrVr10RkQKXxmaC6IZvN4sw+ePAg9pbIQ/2zsrIiFy9eFBGRq1evGsaSzku32zX0ie6/Xq9nXGJ0HLxe79BLskNpOeKII4444ogjj7zsivD4fD6Znp4WkYG1qBZ0OBwGMvD555/LhQsXRETkzp07sLaOHj0qx48fF5HBrePTTz8VEZEf/ehHsL6/+93vyre+9S0RGTginTp1SkREvvjiC8DHLpdLnnjiCRERefzxx2HxbW5uAml5//33ZWZmRkQGFqWiK6OI9YbPULXC2RcvXpSzZ8+KyAChYuhZ5fXXX8ct6/Tp0/IHf/AHIiLy4osvGjdS67tFBrcv7Qt/3+12DfqPIVK+RQ0Tn89n0EyMIKlDX6fTgXWs7dXfsHXPN3x+pn4fDAYNWoIRDUa3GMlhhJCRCOt47STFYhFrKhwOAz4Oh8PoC98K2u025jkcDgPq3dzcBFIYCoVwIy6Xy5jrZrOJm1IsFsPtUeQhlN9oNGxpAo/Hg/6Ni+7wfDPK1G630Uev12tQWjr2/X4fa4cRiWg0ivazw3OlUsG6YLqQHbbZOZad/fUd+t9xkMhoNIo5z2azuOXmcjnQ2nNzc0B1RB7SFN1uF+Pf7XaBbORyOaA6169fl+vXr4vIgM7VOY1Go7gVf+1rXzP6qGOSzWahYwqFgjG2vAaGCVOK1n3At3l2NuXf6Pyys386ncb4RKNR/KbdbmMet7e3oc82NzelWCziXUy5cztVeN8Pk9XVVcORXPdQtVrFOM3OzhrIhp1zMq93XoMiAoQnl8shaCSZTAKR6Ha76He5XMZeYV2v/y5iUsSjSCwWw1k4OzuLM69arYoGT4TDYcMxW9djuVyGrtra2oLu8Xq9WMuBQABrn53KWef6fD4j0EXHJ5lMYkwWFxfBymxsbBhozDD5p3/6J1lcXBSRwZ549tlnRWSwD1SXbGxsGOtE9e7Jkyfl6aefxvjovNfrdcM+UGovHo8DNapWq8ba5L7vhC4zDTqMQt/V4AkEAvAoP3XqFBoYDAaxSQqFArzFK5UKorGOHz8OPxte1OVyGcbP4cOHQXvt3bvX8GT/xS9+ISKDTasQ8/T0tMEDK9x89+5dTHg+nx+LU6/Vagbfrwttc3NT/vu//1tERD788EMszGq1avD97Beim/u//uu/ZGlpSUREXnvtNfn+978vIoMIFqYf9L2ZTAYKq9VqYeEzPBwIBLCQU6nUWAaP1ceFIWymqxg2VqXpdrtxoMbjcXzPSpDpEPa94PeyguHDvtvtwhBhOop5+2FSKBSMcWLFwcJjr7/nzcOUE9N6Vn8GHTM2yKy/1/HgQ/9/hy7g6CorpaXi8/mgULh9HOnocrkwn6FQCMaD3+/Hs/igZ5iYKadAIGAcgnZ+Nf8TikTftXfvXly2VlZWsD8WFhZw2LAhzxFJd+/ehWFz48YNGDx37twRjTxhoyUQCOB71gFHjhyBkfPkk0/ikC6VStgriURirL24k/BY8WemnDiyLxwOY0zS6bRxQOrcMa0ZCoWwNhqNhkEp8yFqZ8Syrhoma2trhk5UqogNHj64o9GooX/5sqTicrkwt4VCAQZPu93GXrf6cPE5ob4ik5OT+N7n86GdpVJpLB+eeDyO8U4mk0Z0oPaR6apIJII1cu3aNfinXr16FfuvXC6jPZFIBLr+9OnTcJUIBoOG/xJTe2rsHT58GOf07Ows3js9PW24iQyTDz74QG7evCkiA93w27/92yIyAD7UEPriiy9g4Ik8pMCOHj2Kyz/vv3q9jradPHkStJff70c7L168iPObI5at/q+6rpnGt9KpduJQWo444ogjjjjiyCMvQxEeRWwOHToEq42difx+vxEpoRbc9PS0EfE0MTEhIoObkjpzVSoVwKz1eh2W8v79++F4XKvVYDkGAgGDXmFKhSMDxoEnG40G+uXz+WBBnzt3Tt555x0RGThnKVowOzsL6C4Wi+HGWyqVkN+mWq3K7du3RUTkhz/8IW4Yf/qnfwrKhB30zpw5g74HAgED7tU+BoNB3GYikQi89UcRa2SL3e2x3W4bUL7OVzqdBnQ+MzODdjJd1e/3DSdOtdCr1SpQrwcPHhiRS4rqiIjhOK3tHMUBTWVzcxNtD4VCuH0xNdfv9w3HaoZ3+Z1Mzekzu92uccPk52hfQ6GQgaLwLdvu1sH01ihizeekn9lhkdEtr9cLhCcUChmop10kjBUO1vHhNjJ8zDlTmH7kz+NSBdzHqakpODsyEjU5OWnr0Nlut7F3b9y4AQr68uXLQIIZze31euhLp9PBfrp48aKBECq6/MILL+BzvV43UCy9CY8iVvSGUR1G7RjNY2dp/X0gEABiE4lEDDSTEVOmcdlRX8fQmpNHx4QRnXGQusnJSUMXKJWztbWFz+vr65jbZDKJd1qDFhi1VXTi1q1b0PXZbBa6OxKJGI6vOjZutxvocqPRAALDUbhWtHKYNJtNtIHbrM/S99pFXVlpemUOisUi9E06nUb7mZZkp2WOmLRSjuwiwNGT40RMKooqMkBMFaVZWFiAw3YymcTZ1u12cTZwlPTHH38MuqpWq+E3/X5fvv71r4uIyBNPPAG0anl5GeudaXa/34/vOdcaR/+OIrsaPJFIBA/m8EgO3eSDKRKJYAGm02mDBtBDc25uDputXC4bC0cXYDqdhmFQKBSMCBD9fa/Xw3N4wXKo3CgSDAaxEYPBICb23Llz4Fp9Ph/e9fLLL4OGm56exmLM5XLy3nvvichgklUJVqtVPMcaeqp9YX8nkYfGm3WBah9rtZr87Gc/G7mP0WjUOAgZNrbziA+Hw6AXT58+DeNzz549tv4c3W7XoD10wxUKBVB7zWYTBiH7MXAoOPfZmrRsN9na2jLoVp2rWCwGQ6XVahnREbpJOPpG+y4yoCo4FF3XyL1796CMgsGgEd5uF+3Hxk+r1TKMhHGEE+VZQ9FZqWkfOXyXfdNYEbPBw2uNE0+ygcyHICdd5L70+30jqmicyBCmGcLhMIzuZDKJd/GFgKlDr9eLQ+Xq1atQsg8ePMAedblcRlSXrkE2DO7fvy8ff/yxiAwuXseOHRORgbGvl79erwcK7NNPP8WlbRSx+qOwkcNG1E70JUc96brlKLx+v4/54HQRjUbDiNTU+Q6FQrjARSIR45Kh76rX6yOv18OHDxvuC6rHmTasVqugCvlg5bFhqrnZbOIC+fnnn2Nd7N+/3/DXY18kXSNMl/D4dbtdw0hQPaG6YzcpFAowosvlMvrA+8/j8RjzqfPg8XhgJLB+q1QqaHMsFjN8DDlamC8W7MOjf9vr9bDemXplH5hRhNdXsViEwRkMBmG09Ho9GCrpdBrPj8fjmPdjx45h3vP5POi869ev48ybn5+H+0sikcBv2Jhhg5uTjLZaLfQ3HA4blL6dOJSWI4444ogjjjjyyMuuCE+n04EFd/PmTUQ/zc7O4rbHiZHK5TIs5X6/D2vL4/HghuzxePAba84TfSYnu2s0Gkb8vVp6TIkwfN9sNkemQkTMW2u1WkV02O3bt42U/eql/mu/9muA7BhJymaz8tprr4nIAP5+/fXXMYY/+MEP8Bumb9hRjhONsaOWXTTOvXv34FD953/+50P7aBf9IWLeSDgCJBKJwEI/ePAgnOYSiQToyG63i/ZwdAX3ye/3Y/10u13DIVmFHQz535jeGia8FrRf/Hz9jj37OT26/j0n4otGo/h9sVgE5XHr1i2MTSwWw7xZEwNyoj8V/jwOvCwyuNXwLV5vUz6fz0h4yZQErylO36/7MhwO20Yk8ZwwwsN7lPcir1nef7VabSyEh+eRqRa+dTMiFwwGjVsr59tR1IXRR7/fj/nlm3C73TbWie5RTnCWTCaBJrTbbayHjY0N6IxRhKOxOp2OEbXHgQV21AVH59XrdUSzMF2rzxL5MqWl48bRkIyGRKNRAz3j5G6j7kWPxwP0Y3NzE8i+3+8Hvb25uYn8MNPT08Ycci4r/X5rawu3/jt37oBSmZiYMPrN+YcYybFzSGbdN06AhLZfqah8Pm/kmVHhc5GpJW2TvpeTRPI+1mdGIhG0v1wuYwy5ZAOfE6VSCTo6l8sZASfjIDy8Bhk1Z73FOeY6nQ6Qq83NTZyRHMXGVGCxWLR1X+AcZto3q7A7Czs2W9tnJ7saPLlcDvCu1+s14EmGOPUllUoF4dXnz583EmOpAlpdXYVi4jok1gRD+pu1tTXQTO1220gypBuSqRnr5h8mvV4Pg7e1tYXoDs4cmUgkkDU6FAqBKuh0Okb4ri6u5557Tp5//nn0RaNN2Hel3+/j99YIJvZ9UvH7/Ti03n///bF8eCqVilH3iJUp+4JwtkuOfuAoDqY4OXyewz118xWLRYyVtQ120Rj6LJVRlRD7n3CmX6v/CYe8cmI6/T37Qvh8PszV5uYmIhGXl5dxIMZiMVBpsVjMMNiZ0lLhLLvjCoeh8ryxcWI1HnmeeRzYAGAq0u457GvEdBg/n6PYrAfrOKGwfHGxpmTQdXT9+nUcBjMzM4aBrAqX63yxIT85OYn5qtVqyG6+tbVl0JF6YOTzeTwznU5DH6yvr4MyO3/+PHTGKGKNuuJUDWzwcOSg6jy/34/xLxaLGHP2/wiHw+ivx+MxMmmr8CWSPzPNwzToOH4SPP+FQgEUYiQSwdx+9tlncvr0afyeaSDWNTr2y8vL8sknn4jIIEro8OHDeJ9dZKT2S2Sgd/Si7nK5MLfZbNao2TSOwVMsFnGeVSqVHaOC7fYf68pGowEDptVqGRFe+pmz5LPhx3UQmZ4tl8voVyQSwdkjYm887CQcLTwzMwPDlVOZ9Ho9Y371TDp79izGJJfLoZ2xWMwAMljf6PhzzUP2PbX6V3KkqV3KlZ3EobQcccQRRxxxxJFHXnZFeLa2tnCzmpubM6AmtV4DgYARuaOVyv/zP/8Tv43H47ghv//++4A8Dx06ZNBhaslypdmlpSWkoH7qqafg3JROp2G9cj0Tn883lkMoQ5uNRgMUDCMhqVTKSC7G1IhauOx97/f7DUc2rhyrbeN2clJBax4bfqdWbv7ggw+M5EzDhMfDmgxQ3+X3+43bFTsM6hrg2mccPcJJ7nK5HNrNVaWLxaKRhEwtd3Yq5DFvt9sjowPsYFyv13EDsaJTnFZenfA4BwfXfeHkZYVCAbfNra0ttDESiRjQM0eG8O34/0SUlrWuG9OefFOyK59h/Y1dPa/dnmlXCoGpFu6vtfzEOH3k57daLYxbs9nEOjp//jzGPJVK4abXbDYxp5zThBFfpmd1LYo8pMJEzKrSrP96vYcV6m/cuIE8YVeuXBmrLAGLld6yi9gSeUj9BwIB6BJGTMvlMtYtI8eRSATzwrrHmj+JaTV2Vtf+sqvCMOl0Ohg/puE5/1e5XMbeYiSBUcZWq4Vgj+vXr2NsksmkUQ7Fuia1Hzz/TN9xkAZX4h61fyIDfaD6t1qtGnmHtC9McVuDGBR9ajabWId+vx8o3cTEBBCVcDiMecvlckAlc7kcxpmfv729jfW+vr5uoHfjlnlR5/B9+/bh/CsUCmhDvV7H9+FwGPP1zjvvIECl3++D/nO5XPh9LBbDuuYcWhyByhQuCydmtNZcHJZPaagPDytuLubG0SlMA+igXrlyBUYOH/oPHjwwathwA3UTrK6uGtlvlSa7dOkSFkI0GgVUeebMGQzA0aNHh3pqs/DmyOfzWERMjUxOTuK9XFCQIx/K5TIMP2s0CytuLtzIPi0M3zMHzz4Zavh98sknhsIeJhxizX4bfr8fm4/pina7bVCKDDHqomYImakupjFyuRwSU3F2Vw6RZR7bCuWP6jcQDofxdxz9xona9PkiA+WvbanVauhfMpk0IuF0E3KxxXa7DWM8FAohmm12dtaIkNLPnOCON7P2cVSp1+uGgaFiNVTsns2RYmzA7BSqai1qqWL9vNNz7OqIjSKsbzqdDpQ1p69ot9tIhnr8+HHs+1qthj3BzxF5mK11YWEBBk+pVAIEf/36dWNfqrDhFA6H0Z67d+8aUZjj9JMLajK1y/421sKsvLbtkl6yD1K32zUS3qku4UgrTq4Yi8WMYo1qWFarVejCXC5n0O67CadwaLVaGDOrvwr7Jmq7eGwajQbm/NKlSwa9wgkMeZ1bffl0LNn1gSNLOfpQdcYowmtzpxQR3BerywWPFUdmaeTa/Pw8PkciEcxDo9HAmuVi0pxpmQ3zWq1m0KfjAAELCwtw4zh8+DDW0Y0bN+CzViwW0R9ej7VaDZFZrVYL7XG5XEbNOh2r1dVVw5VEhddMu902Umiwa4sK02E7iUNpOeKII4444ogjj7zsivBYayexlWp3++bU5P1+H9YoQ4/dbhfOz1NTU3Bs9ng8sGTv3LkDhKfb7eI5b7/9Nqz706dPGxVZ1XqdmJgYy5JlK7JWq8GSbTQa6HssFsPN5+OPP5YbN26IyMAa1ZsBV99Np9PI33Hq1ClYynv27DGgf064ZQc3u91u3ABu376N3DvFYnGsW2UikcDzk8kkxpxzVzC0zWU7lpaWDKpAEZtut2s4Nusz9+/fj+fkcjmgIcVi0XAS5HXFibL45jSqLCwsGNC53vRDoRBoz0gkYiQP1H50Oh05evSoiAyoV85hocjitWvX8LdWyFTfdevWLazx+fl53L6s0RnsFDqOE2G9XjeQNkb+mLpiWoTHkBN4MSKrbbBGBtnRK5zenR2bOSLMmnhw3JpvnKBN9xajDY1GA46bzWbTyEujc+TxeAwEUffQ7OwskgdyBBGjXjw+1tu7tuHevXvQT1xrbhQJBoNGbhGmzHjMVRjFsNISnItJEZt4PG7QDDom7Hjs9XqxXycmJoBMcw2yRqOBz9vb2yMjytboIU76x3mw7Kglpvi4LtXt27flq1/9qogMzgxGk+2QjXA4bMwJoxz6Xh7vZrM5MoKlwvmC2DGc97RdJCUzJawPkskkdNXExAQoSo/Hgz3B+X+KxaKBXNmtEUa7OUp2FPnGN76BM2x2dhZzce7cOeRE4vYXCgXMSywWM1B2XV+RSAS1vQ4dOoT2f/rppwgCyOfzhruMtpnRMGtdLdapw/biUErLGsYs8mVvb058pv/WbrcN7pmzYKqimZ+fh69LvV5HRNiVK1eweP1+P/wnVlZWwB8ePnwYg9dsNvGbRCIxFlXASejY8GB48saNG6CTlpaWDL8HFv3+/v37cvXqVREZ8JmvvPKKiIj8/u//PsL1mIpwu914L2eA5U188+ZNuXTp0pfGdhTxer3YQOFw2OA/OTRXN3G9XgeczIbr+vo6uFmRh+HCqVQK86iKVMQ0eJgj5+ggay0rVvqjHiTHjh1DP7iWWrlcBg/NPkdbW1vYkFy3JhQKAYpdW1sDzLq8vIy5mpqawoHOobZ3797FGBw4cMBQxDvJOAclJy3ktWONzNoptJkPbk7kyfQK/61d+62QPRdItUtqV6/XoaRGEY6e44MhmUxiXrg+E1+8OAKSo1Y4apAjOri2lDXKg/Uch2bb0ZHj+u+wT0av1zPoZYbs7dwHRB7qWk6+ls1mQa1OTU3h+06nY0SUcoJYpUymp6eNWlBq5HBttVKpNHIh31wuZ+wt7UepVIKODoVCRuJaTnWgY5PP56F3Op0OzoxEImEk9NPPVnqLfSLt/OnYcORoqVGk0WgYKVe0v3v27DEiV/V762WFaRpOnKgRhPPz81gXnFiPE/d5vV4jYtbOb87n88H4Zf+sUeSFF4bXdqwAACAASURBVF4w2qaG1sbGhnFmMCWrfWTfpEAggNpbJ0+ehBtKNpvFGbO0tASDyhrRxhcXNowZWOEo0mHiUFqOOOKII4444sgjL7siPHzLYuuYHVy5pg7fIti7mm9QPp9P9u3bJyIDBzS9nX722Wfy5ptvisiAQlAL3efz4b3BYNCwWBVFuXLlCpCEZ555BingRxG+nTJyxTlBlpaW8Jl/HwwGjZuK3ga55lc+n5ef/OQnIjLIO/Ttb39bRAYwIY+RXR0jr9cL59pf/OIXGBOuID+KNBoNW9SLo6JEHt56OGIun88Dst3a2gKU7/V6cQPgpHiFQgHjc//+fSNfhQojBYxuMcpkHZPd5NChQ3Ck8/l8xs2K67bp7Wh5eRnvyWQyhsO4jvft27eBKpRKJay72dlZ3HYY9hd56HTPVctzuZyRM4dvmONQWiw7JTPk24410odRC0aKGFHjm7Md+sTOyYzSMaXV6/Xw/Th0lv6eSx7orT4WixkooJ2jb6PRwDqq1WpGf5mKsnNs7na72Fvs6Mu5aFhCoRCQzEwmM1auoWg0aqBYui84sICpXe2DiJmLKRwOo8bgvn37cIvOZrNYz6VSCX3nZG2xWAxjOzExged4vV4jTT/r8lGRus3NTfRvenoa+v3u3bty+fJltFF1CpcD8Pl82H+rq6uISs1kMkDGfT6fgVTY0UZMzXm9Xjyf55yjCWu1mrGPh4k17xevI0bpdhJOSqrIN9fPisViRuSw6tC1tTXb/HdMNTOdy0FE4+b/euONN1BC6fDhw0gU+dJLL6FtH330EdYvI2/NZtOgLxVBf/zxx4HwdLtdBDVxrj1GPWu1Gp7DeeiYnqvX60YtuGEy1IfHLnul/ps2hEMr1QDo9Xr47HK50MB9+/bJqVOnRGQAv+pk/uxnP5OPPvpIRAaHhFIwXq8XmTVffPFFcLnBYFDOnTsnIiL//u//jsErFovgQtVXaJhwciN9b61WM8IN2e9BPc2PHz8OysTv92Ozrq+vY7M+ePAA3//jP/4jxu173/ue0T47CoHr4liLUz7xxBMj9U37ogu+2WwaCpf9ZzhagjeILiTOcsu+DuFw2KAmtb+cuM1KaTEEa5fMjmmbYZJMJqEsisWikVSN/TrUWLt9+7YRLaBz0mw2Aa3evHkT/eAEg9PT09ic6+vrRkQd89aq0K2U8DiHIwv72zB8b627ZMflW3luVhy8d+3WIM8BR0rwgcF+RCxsUI0iTOdms1mMOe8TjvqoVqvoI+9dfq/P54ORs7y8jPllKoKNpXQ6baxV9rfQQ2Xv3r3QYZlMZqzDJJ1OG9FEnHlWP3OmeTYCObUCU1oTExO4AOnhImKGLjMtxYnbuO4RR5Tu2bMHRlEoFBq5XlihUIBOzGQysry8LCKDIq7qjvDkk0/CyEokEjjU+IJy9+5d6PQTJ06AsmP3CKaH3G43+sTRtoFAwKDZORqP/UrH8eHhCxvTwiJie4m11nfkvcKh6JzKRPu4tbUFfbOxsWHobqamVdgY4+zKHPI/irz99tuG0aLjf/z4ccMoVT9H7p81sasauiKCfXPixAm4pORyOTxzbW3NoKbZdYPpXAZixhGH0nLEEUccccQRRx552dXk4wgT9mrniC2v12tYYUzHcH0dvfFyZXCv1ysXLlwQkUE6arXu2fEqHA4j78bp06cBiW1vb8PKK5fLiJy6c+cOKIdRaJ9wOIwbIFvBXK2brenHHntMvvnNb4qIyJEjRwx4Va3NSCQCT/Yf/ehHcDZeXl6W//iP/xCRAUz43HPPob92yZY4adqrr74K5OratWu4/Y4i1WrVto4OQ7DRaNRI3sfIiP4tp7lPp9OgDufm5kD5eL1eICm3bt0C5Mm3C0ZvuGZPsVjE7TASiYxUuVj7obc7vgVvbW0ZXv6aQ2h9fR3viUajxu+VxmLn7Pn5efR1cnISY9Dr9Yy8RIw8cK4QFb6NjOOwrO3kpGx25QkY3mXHV55nRpnq9fqXypeIDMZeb7xch8uawJJrr3Fklgq3bRRhhMfj8QDB4GrpvHY4Ssfv9wM1iMVioC/591tbW8ifE41GsTYZrWI9l8lkgH5w/qonnngC47yysjJW9EsymTTKy+ia4SjCdrtt3JgZ7eHbr84d1wiLx+PGWHFeHUZGGNnVz36/39jfOv5erxe6eZhsb29Dv6dSKXnvvfdEZIDY6JjNz88bCTvtokM3NjaADk9NTRk1uRjVYSpV1w6XP2DKzJq/iud8nHW6U0kWa4QWRzHyHOr68nq9GIfZ2VmMN9f3W19fhz4tl8sGassUDjvx8p6wKxc0ily/fh3jlkgk4OQ+OzuLXFZ3794F+pTP540ktjo+7XbbyMmjey6TyeA5IgKEp16vI/lvr/ewyjyvdy6fEgwGjXEeJrsaPAwN80ByeB/zh1YomTetGh/PPPMMDpuVlRV59913RWQAj9mFuUajUfDTBw4cMLzg+dDk6JFxsmY2m01DcXOEFCdf+8Y3viEiA6NLYWNryKgq/VgsJmfOnMHnv/zLvxSRwQJRiPfKlSvy9NNPf6k9nKGTk/tls1kUMJ2enh4rqkCfK2JCnuxrxBQI14LiQ7rdbhtQrm6gdDqNMeQQVqYKmG5h4QOJfb3GyQzKiQ+txfb04OPD0ev1AqKdmJjAplpeXsaB+ODBA/gNzMzMGMad9vvevXtQygzjNhoNg2phmo6V4jhK1prYj/lyDuXm8VZhZWEtYMp7iH9vJ1ZYnP0VeK8w1TUOpcXGL1OmU1NTmAuPxwMjhP37fD4fLgF79+6Fkq1Wq2gz6xuu/cP7uFarYS1NT08bNIP2kQsiplKpseiQeDxuHAZq1EejUSOEmA9F9lPiOkx6eJRKJcNA5QSoOia8TprNJgwY9u1hqiuVSoEKcrlcBlW2mxw9ehRzVSgUcHG4c+cO3BEOHjxoJERUKrJUKsHgefDgAc6J48ePG5dbnYd4PI4x297eNnxJ2VDVd7H7BUfArq+vj6VPOSrKSjXbRdpZw6W5JpTO1dzcHIw6n88HnyL2m1SaTuTL0ZkcYs/1xbg949SYdLvdcuXKFREZGKg6d4uLi3Axeeyxx3CZtyac1Pfy+r1//z505MLCAsLeDxw4IL/yK78iIgPDh41eLjTOLjJ2hhyfHzv2a+QRcMQRRxxxxBFHHPn/VIbm4VGxOl/apc22eq9zaYZnnnlGRAbWvd6K3333XViRlUoFFlqv1wONEo/HAaeFw2HcalqtluEYxcnXxnFkYqdAq7WoN4PJyUkkD5yenjaQLo5s4rwVaqE//vjj8qu/+qsiIvIv//IvgO6WlpZg+TINY3XitfNMFxleM4SFb8Jer9c2gROjKZw6XcSEanWO0uk0orSY1tzc3ISFXqvVjPbzOLM1budEzennh0kymTTKQOitXOQhVNpoNAAfT05OAjXMZDJAdS5evIjIv42NDaCS09PT+LywsGAkOONaXYwsccI3uygOa5mJYWK9nXGZAF2zjPBYHSX5VqnoRCgUwlro9/vGrUnns9lsGunjGfmzS1RorQ81zl5kZKzb7RqJPHW+mAbg9P1cGXpxcRHVtTmR2b1794x0/DpWlUrFoDf0Brtv3z7ML9eiYvQpFArtWC3bTmKxGOay1WoB4YlEIkZpCbtkklZknUs1aL9YF3IkV6lUAqrTbDahn/jmz7mVYrEY0J5MJjNylNbBgwcxlnfv3pUPP/xQRAbr9+tf/7qIDBxWtd+dTge0MFN2165dQ0JQpkhY53IUHZcvajabRlJGO2d/rhVWKpVGRrBETKTWSh0zOs9zyHW7+IzRz3v37kUb/H4/5ofzEW1ubhpUOVPczGrsFGE5zpkh8jCXUaVSAQLG9bDm5uagazc3N42gB0YrGWHTs//mzZugw+bn56Ffjx49Chq0VCoZOaL0+aznOGhjFER51xHgP7YmI+OXcBI0Vu4qk5OTiCqKx+MovPfOO++AcmBu1prsjCfZLiKFaYNGozGWAtre3jZqQulhwAZVKpXChucQTTYEeGJrtRomPBaLgT7h4nLlchmTPzExYRsuWa/Xjc3EyRvt6sbsJKFQCO1hmoypCKswJ6yfQ6EQoOJMJmMUVFXDIpfLGaG/XC/MTjFoO6yfx8kMGgwG0a5+v4+NxPWPWq0WNs+xY8cAl7daLfh/XbhwAVEH7K80Oztr+PxovzlShnnlYrGIwyUWixm0IRsG44Slc+01ppd53XHWWjac2RjjrNiBQMDIAKtjxQZpIBAwDhgVa0I8Frs6XKMIRyUyNL+wsCCnT58WkcFeVz8+Dt/lg23//v34DadYYCrH7XZj7tiYDIVCiB45duyYQevonrPSreNEaUWjUWN/6xqLx+NGJnIVjthpNpugf/x+v+GvxYVEtZ2RSARrlYtubm9vw0DK5/O2dGcwGISxZI2W2U2CwSASdr755pug8F988UVQ+Nls1vCt0zXLdaMuXLiAiB6+WFgvZrquO50OxoALAgeDQaN/+rndbsNI2NrawqV6FGEjhy/2VtrKjv5lY5YTQM7Pz2O8t7a2DGNMjR/2t2L/JetFhxNq6v7mZI+jCK9NTofAPqtsvBcKBXwOBoNof6PRMFxGVMcUCgXsxYmJCazNiYkJrFmP52HtM7/fb7ibaN9ZD41yXjiUliOOOOKII4448sjL0MSDikh4vV7jpsEWFlua7Cmvt+6TJ08i2WAul0O+nZWVFVh/DK1mMhnbhG4iYuTD0baFQiHcdlwu11gIDzubRqNRwIp8A+cbHSNabNFz/Rav1wsL10rL6M2Ka28xvM40ANMS1qiCcar78i3dCgnzODBMyFEITIfojTeRSBhjzunP7fKYcMQOf3a73baV4rk9w4RRN4Z3e72ecXPQ5Fn79u1D/9bX1+HgWigUMIeZTAbUCTtlM/Jz8OBBjEG1WgWylcvljNpGdlEl4yYd9Hq9RtQgO1/a1WDiGybTmBz9xjfAVqtlS8/xc62QPaNV7PzMCM84MLrf7zdQFB3zffv2YcxFBMhMLBZD+7nu3+LiIurscd20er1uIH7a/lAohPbv378f0SNTU1NG9W6ORNO/LRQKiMgcRVhPcAQRU1pWKlDfy2VBuGZWKpUyInBYlzDKZ0fDMULBFefL5bJRz2lUtC6fz8NN4cKFC3D2/973voe9wlForGdbrRacnBuNBnRNNBrFfrWiw7zGtR+lUgloCedssdac0vEeN4JJxMyxw6Ux7NA+Rlv5nPP5fEZJEG0DU46FQsGogaZizXvDekXHwePxGCWFxukjBwKxY7s1UlOFkdFsNos9cePGDUM/cZQhB15w1Xh13p6cnDSCoHgt251P2u7dZFdtZB1Uu0R8XJOm3+9jYDweDxb7yZMn8beffvopKASRh5M4Pz8PRfPYY4/Br2Jtbc0IWeOwcT3Y2DeC2zmKcDbpTqdj1A9RyimXy4F6S6VSRqicKiNWCK1WC3Cdx+OBTwtHNgUCAeNv2BdIE24lk0lsXDa6xumf/p4PSzZsODyVn6vjHIvFjEzXvND4kOPFy9lgmabkcHu7Tcwc+zg+PPV6HXN1+fJlOX/+vIgMjBkdv2PHjgFSn5ychGJfW1tDG44ePQoqZO/evYgiCAQC6Eez2YTB8JWvfAWHzoULF4x6MGpcWaOiVMZJrGgVVnBM87JRzIcjXwgikYgBE3NWZLsMzLzueA6tVJqK9XAcp49MsXU6HbQ5FApBcXMECB+W3N+ZmRmkstjY2ECbNzY2sP/cbrcRzaKU5enTp+Ev1G63QbEwjcg19/L5PGibUYSjV7m//DkUChlGi/pJeDweGOF79uzB2gsGg5ivYDBorFU7SplTHOi7RUyjgaO6wuHwyAb67du3QSnPzs4i+mZxcdGIXNT312o1zMPKygoo5RMnTuD8YF250/piXczPZLqSjXTO0p3NZg1fkWEyMzODdcTRuXyxLBQKWI9bW1sGPah9/JM/+RP5jd/4DREZ6CRda9euXYPv09WrV3GZZBqzWq0ahrDuD32HyOByrWshmUyOBQR4vV7jQmBHJ9VqNZzB+/btg4/W/v37Ybi+++678KdjQ5f3Qb/ft02nwXPa7XYN+prD0pleHKZvHErLEUccccQRRxx55GVoHh614AqFglGOnm/3agmyRRaPx3FDnp6ehvW6tLQExCYSiRj5LPRWtri4iPd+9tlnyENQKpVgWXc6HVh84XDY8NYfJ/qFo4d6vZ7hlKttvn//PsrXz83NwRpli5IpPx0LkQGCoJRJo9HALW5hYcGoC6ZW/E9/+lN56623RGRQt+Q73/kOns/OxuM4SrIDaCAQMCgkFWulci4dohY0o3nc31qtZtSX4rw6jNgwgsTt59ubXUK0YVIoFOAouba2hrXj9XoBi3Opgvn5eVlaWhKRwU1Zx0ZRGZHBelR6c2JiwohS0X6XSiVQlFx6IBqNYo1YS4Vwn8fNUcOIip3TsojY0kmMfnCEEe+V3fYNRwMxJcQUjJ0T7061qHYTbbPb7UY7o9Eo2s9JNKvVqhFAoOt0YmIClBYnDdU1ou/R9RUMBuXkyZMiIvLyyy/L4cOH8UwVhtRLpRLW2MbGBpC9UYTRQo6KYkfxYDAIFJlrRCWTSSCQnAtIx0Lky1Q835B1bFOplKFLeK8resmoyjiRdr/85S+xJ1555RU4m7fbbaPsECNzOg/r6+tAtzmwwOpsa5evjRNkcnBIIBAwIrP0XeVyGfs4m80CgRlFGH22Bh9wkk7dH1wjMBgMArFLJpNG4IeeN6urqzjz2PWB0Qym0lhPWikzlXFZgWg0irXAiBbXgrt37x4+nzhxAnO9uLiIIKVmsym//OUvMTZKV2UyGehm1hG1Ws0YN50jK6XMkb36G6a6dpJdDR5rGB/7k7C3O3+vn7PZrBw8eBCfdcLn5ubk13/910VkYBTpYg4EAnLixAkRGQzwp59+KiKDTaA1WPL5PJJaxeNxTIjP5zPCksdJsMRGCnO5zHP3+30sxna7jTHhcFbrotY23Lp1CxBmuVzGAl9YWDCoNIWBf/7znyOZk9frRdbSY8eOQal5vV4j9HqUPrKvgPaRDVfetBzZ0Ol0MM4cis51dx48eAADol6vY3xYUTHUznQYUzvWqLdRfXi2t7dxKGQyGfDi2WwWm2phYQEREel0GvNQKBRAV+ZyOYwxJwXjLLtut9sohGpHGbBRac2uPA7fzMLGCNMrvBetlw+WYe/a7ZJgd+BZKUo72mtcg6fZbBqZ2tn4sYv+ZCXYbreNaEKmMnXtq4+Bij4zkUhArxw/fhxrw+oPqONQKBQMo4vXwzCx+lZxgjb2mVDhMOCZmRn4QnICzHa7DQqd/Sg5SpINnng8Dv3BFxqXy2X4CHFk1Kh7sdVqIaz/ySefRKoAzZ6rv2F/DB2/TqcDqjmRSBj+o7y+mNLkemic+JUjjNjHVH9frVYNym5cYdcKLhrN4djaHp7bVCqFMZmfnzeiJJWWtyZvZbqSk4zqXrGePXwR1d9zf0eRXq8HPXr48GFjvWjbbty4AV+jiYkJzIsCGta+F4tFrOVIJGLoVzb2dBw4waDV/YIvXvrecDg81GfQobQcccQRRxxxxJFHXoZSWnbOQSxWGJSTDSolkEgkYNlNTEzAAuU6Km63GxYlJ2Wr1+vy+eefi8jAIVURhnQ6jRsA17xh634UYWcvj8eDNpw8eVI+/vhjtE1h67W1NfSX89gwAsaRD5999hnQG7/fj9vPwsKCcWvR53CejkuXLsmPf/xjERH53d/9XfTRGqkwTDiXB+ca4ugHTjDHUSKcqG5mZgZzmkqlQGPdu3cPybEY6Wg2m1/KqSQyWFeMULBD9U7OibtJIBAw6i6p9d/r9YzkaRxxo9Ltdo3buv4b3755DHiuOp0Ofm+NcNHvo9Gogbqwc/04kVpM2eyU7HOniBG+ETGKws7J1pxJds7sjFxysklrMIHdPI/aR75tM5qjN71oNIobINMhHOlRr9cx/olEAgnsjh8/bjyToyR1foPBoFGigNPZM82kc5FKpcZCW3ldWfO52KFsjLpEIhGMTyAQMJAcpvR1/DmxKEfqJZNJIFpM+ej7dBw4kGJUdODUqVNwZUilUlh3qVQKSA7nafH5fNAjjBTynt6p3A6vWc6bxgk1GeHhZIP1eh2/LxaL0O/qFL6bRCIRgxZmZ3YdP3ZCT6fToOc4waD1DOAIJj3bgsGgkWyX3Q7YOZ33JbMOmtQxGo0CjVGUcDcJBAKghRcXF9HfQqEAh+oPPvjAQGMUpbly5QqoyZs3bxr0vo4P59sJBAJgcdbX142AFs7/wzQo6z9G31Uf7MTy7GrwsHIUeaiArKFpzBnqIpqYmDAWLE8+J4JiyFW/r9VqxoLVELef//zn4ACffvppLKIzZ84gsmLv3r0jLVqV+fl528FZX1/HwtzY2EDU2KVLl2zrKvH4VCoVOXv2LNqsBk86nUbUwqFDh4zDQNv81FNPgfNcX1+XN954Q0QGtJHW81pcXBypboiK2+0GVcNcOnvf88FfqVSw0DKZjBHarYrS7XZjXra2tvC37F/EUCsrbqZerOGhdhFEwySdTgMyDwaDMIpFHoYwp9Npo8CoGie8kTiRFqcKKBaLaK/b7TZCRpVKWF9fN2ozqeE8MzNjJAy0q7E1itj56ehnbbPH47HNWs2+dfl83vBdYeNUFSv76nCY+U7+QmzIWbN3j9PHVCplm+BM2yFiQuTW2nfaZq7hxheyeDxuGDNMxdsZkGyQWJO+6b6fm5sbi9LS91k/83dsrPIe0n8TMRP2WdMR8L5hHy2mq5Ty44Sv1vnS8SyXyyPXmnr22WdxkHGUmM/nM7Irs9HC1BlHS7HPGq8v7q+OPftRMR3XbrfRDz5v2Ei4c+fOWEn5pqamjAs2p1ZR/ZXNZqErOQv4gQMHYIDv3bsX7Wm324YPkj6TI105GouN/VqthrHlRKcejwf6iZM6jiLf+c535MUXXxSRgd7XNiwvL+NsW1pawr5cX183qFEtCn7p0iXo46mpKRhRhw8fxly3Wi3o7/v37xvzruPJdRzZmOSLAtPPO50dDqXliCOOOOKII4488rIrwuP1emE5ZjIZw/pWyy6VSsFiPXLkCCKSpqengVpwHgd2bG6327BS2RmqXq/Dws1kMrg5X7hwAU6oIgJP8BMnTsiRI0dEZGD9Kcw2ivAtkRMGTkxMAJp98OABLNwrV64A4ZmcnAQU3ul0JJfLicgA1fnnf/5nERk4Yem4PfXUU/L888+LyABiZDhe5Wtf+xqotLfeegsw+kcffYQok8OHD4Ma+/a3vz20j+ywyDQR3ySYXnG5XJj3bDZr3GY4saSiRgx3Wx1qrc/dTTgxn8jo1cSXlpbgeJzP5zFmTMEobC4yWHeK0nCJj2KxiPdHIhGgifV63Uj0qO/iFP2lUgnwcTAYNBz3uP875eUZJlakZKdbOScDZMibq2nr7xnhazabtg6gOzldc+QXw+hWamac3B+cJM7v9xsOydxvppm0ffV63ahxp8LRXvzMRqOB33NCOh4HTmfPFIjH4zHK3Yw7j5yIknMlcckJdmDVNTwxMWGUn2Aag+sM6fi3Wi3o1EqlYiSn0xsyU4iNRgN93ymh5TCZm5vD2q9Wq0bdKNWtVgpG29VsNqErp6amjGhYzh3Ga82aA0rErEjP/eDIRS5FkkqlxnJczmaz0I/cfqY6k8kkGIJMJgPac+/evUB+GA0vl8vQUUzPxuNxI2koO3tz1CD/hveLojrhcPhLqOlu8tprrwGZ6Xa7iBo7f/48zieRh3vz/Pnz8sILL4jIgH1RxK1UKuH8OH78OCK5jhw5gr7fu3cPbiuM8PC+ZHo5HA4b88sll4btxV0NnmQyadSVUZiekxLF43GESh4/ftyoE8LhrDrYrVbL8KXgEGZVjlwXJRAI4Jnr6+vy9ttvi8ggwdVLL70kIoNER/qcer2OqCgd3GGi7SwWi3jX/Pw8KKStrS1MyObmpvzkJz8RkcFkq1G3srIily9fxu+Vk+x0Ogh3fvXVV0G9sZ8ER0FkMhn5nd/5HREZKAzNSl0ul7GItre35eLFiyIi8rd/+7cj9dGu5gxn0ORoqWg0isN+dnYWGzebzRoGhBoN1iSBrGDsorHYk56TELLvENeHGSavv/46oFuOBgkGg1BMHPVRqVSQGKtYLBqZofUgO3DgANa+teYYRxKpwZtMJjFmnPmW6TtrGoBxhDe/NcycLxNMx3BKAO0j+wG0Wi0j+d5OmX7Z50SF53YnJcNrbRSp1+tGQjduD1M/vGZVZ7AhYfWN4gy8egCwQcppBNhQFHkYacgFddl/KRqNGn6Aw4QveezTwJGLbJBvbW2hbRy9U6/XjSSv7Lul47a1tQXjnCNK2X3A7/cbEbcqrJPYX2SU/nHUjOr9SqVirDUOhVep1+vGBZsvonw2cLg808X6+0Qigc/tdttYp+zHpLKwsDBWKpNwOIxDWfWOyGAOdX1xorxQKGToA23z6uoq9Mf29jaM3FarhfUVi8Xw+1gsZmtcsR4PhUJoQ71ex/xzlOwoEovF8JyNjQ0kc33nnXeMPupa297elnPnzonIYGy/9a1vicigeLaugbm5OaPmofr5XLlyBc9fXl62TaHCfotct5DTyljpXDtxKC1HHHHEEUccceSRl6GFbthiUmtxZWUFll0oFAKNVSqVYKV+8sknuJmk02ncTDg/ATtRseWrfy9iJgvz+Xx4/tmzZ1GzhR2P2+02aIa/+qu/GjoADPF7PB6DjtFSBI1GAzeu1dVV3A7/7d/+DVZnPp/HmDAMOT8/j+SBTz/9tBHhwxXMVZrNJqi6P/uzPwM69NZbb2H8u93uWHVROKEfV3Jnp3SOkIpGo7DEU6kUaER2lGMqqFQqGRQL38Y5eoeTVbKzqV3uhHFKhFy+fBmITT6fN0pb6A2KHZIbjQYcyRnmjUajRj0bV4VvzwAAB7NJREFUTjypbcxkMnhmNpvFWG5vb6O909PTcKi3Rp6pjIvw+Hw+o7QLIzyciI0pLXb+03lrNBrYH4yWWJ1Buc0cIWNXXoGdU7XPImI7r7sJP6PRaBgoDNMb3Ea+OTOKyTdPnaNOp4M5LRQKeFYymTSqojMioN+zc681MmecfnI0EX9uNBpGjhWeC21/oVAAbZpMJvG3LpcLc1qpVPB5c3MTqDDn02KnbnZKt86voiq1Ws2oZ7ibrK6uGvSKjiXnhLEivPz/ipax47H2kf9G+8EOyTpX7IDMFBhHqnHCP+3vqNLr9YyzS88Gj8dj5L3RZ7rdbswVVz/f3NxEf5l2drvdmCsuoeT1evF7RroY8eV3eTwenLtcm3IUefPNN6HfNzY25NatWyIyQNB1zDkqzev1ynvvvYe+/9Zv/ZaIDKL2eK3p2rx16xZKsiwtLcFthaurM4JnjQZnXchnv8pOTui77tRWq4WEcqVSSd58800RkS9FJegkc3G+7e1tRBuJmNlX2TOdo7Q4CSFvPE6Ux0nNePBYSY3DqTPUy7wuZ9B8+umnsaDefPNNUFdcS4QjD/x+vzz33HMiIvLiiy/is8/nw1ixomHhbJFHjx4F93vo0CH54IMPRGRA7dn5iOwk1rBO9u1g/xsu4KYGTzqdNkKy9b35fB7Kt1wuG0qFDwbe9LpIOVmlNTSXfThGnceVlRW0JRAIYPyY7uH55AOOjRBWiJxpef/+/UZUC/t1qITDYSTaTKfTRgQQR6Tx53HEGorOY6bC1IM1+zEfKrznVPigt0YM2WWqtYbhMy3IRUjHEW5PvV7H8wOBgOFjoePu8XiMApfcNjWAv/jiCyjWQqGAg6pSqaBfHO7NSSbj8TiSVU5OThrZZvX3fCirX91uwhFYfPlrNptGOgU2OPX7fD5vpLjQvvf7fYxVsVjE4ZbP542wYR2fQqFg+DlyXSv9W2sBy1GjtDhJK9cOrFQqoH8mJyeNsHTV+4lEAvqOs7pbs7Lr+mW/Ks6qzmcMG1RWw1T3yrhRdpxOpVarGWHUqk99Ph90aLPZNN6ttDuHcrMu9ng8eCavcU48yBm42R2E/aA4RUGhUBgrEu1v/uZvjP3EdCvTyDoOa2tr0Enr6+vw+VlcXATVH4/H4ef68ccfA8yw+nRx6gD2GeSabyps8DQaDSOtgZ04lJYjjjjiiCOOOPLIy64Iz8TEBKzR1dVVw7mNLVB2UlNYrl6vG1Y2IwCc5Itvy4w8MDTPN092YmLonL8fB+HhqLFGowFrkZNacV6VV199FZRTPp+HtRsIBGDRP/7448ZtX8eEb26cf4LfxVA2O0fu27cPXvO9Xu9/dCsRMdENTtAVDofRnlQqBcg3HA4bNBbf+rjNDIvz2NpVYLfetBjxYwRvVEfCRCJhIDY6h5zXhz97PB7kxQgGg0a0nN4uOGpwdXUVCBK3d3t7G5EVPLeBQMBwZOUxsEY9jSpMafl8PiP6ya5UgbVOFkPefFPiiCd2rmaElaOldHxqtZpBo+j3nOuEk+ONIjx3XBGZ1ynfMEXEuOVq++v1OhCes2fP4iZZLBZxQ7ZG73CSSYXORQQIz9TUFPZfIpEwqs/rvGs9rt3E7/cb6J8KO11z/S+unJ7L5dDfZrOJPc0ID1MmjPb4fD4jEESfn0wmjRsy092qtzjycZgwLcwJWznajKu7M4KfyWSM+oIcLcfjxKiYSjgcNmgqpqoZaeQgGXYAHmcvBoNB4/kcaMHvZPSRhWtCqVjz8DCqxfXQGL3mSCVr//Qz67ZxnJa3t7e/RA2KmGe51Vmef6/uJlevXsV3fN5XKhWjDAdHW+q4+f1+rMFmswmU3RohyolOh+Wn21UbzczMgFvjmi4cxcG8WSwWwws5AZKIGEaLfmbP8UQiYWwIFT5I/H6/sZk4IRMnXxtHyTKl0263jeR4nIiN28lJERWCnZiYGDoJjUbDSArFRoJdJAwXSePDht87rkSjUcwjhyVz7aJutwuo9d69e7Z+LblczoC/ub87hbMyvWRXJ8nK7Y9K+8zPz+OQ4tBTLpTJhyb7ZgQCASPjqr6TYX89PHVstK8cpdDtdmHwhkIhg4dW+Z8aOyKmwcNzxWuNa8HxumOFzv4/XCBS+yBiXmisvjraX2sdK1Ws1lDSUVMLaJt1jYRCIcN/htcXZ6dl+pRDy5X62djYALzOycs4nLjf7xtRQHzw8IVMdSErbjak//iP/3ikPjL9bpc5V9+t7eQCikopswHDBk+5XDZSJej3TO9ub2/DH7DVamE9WDNy6/PHybTs9/thiFWrVRg8lUoFRVmZdi6XyxiDaDSKOW80Gmi77iuRwfzoIcjRTHxh63Q6tjXl+GBl/8JerzdWyDavU77Yc4JE7iO7MjA12Ov1sHb4/UytW8UuLYSIGGvKTueM6zPY6XRsa5nxXrHWr9M2s7HKiVGZAut0Ovi9y+Uy/NfsQvIrlYpxOee54wvWsOz1DqXliCOOOOKII4488uIa1/JzxBFHHHHEEUcc+f9NHITHEUccccQRRxx55MUxeBxxxBFHHHHEkUdeHIPHEUccccQRRxx55MUxeBxxxBFHHHHEkUdeHIPHEUccccQRRxx55MUxeBxxxBFHHHHEkUde/hfjAJy/hqM4BwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x72 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label for each of the above image: [2 6 7 4 4 0 3 0 7 3]\n"
     ]
    }
   ],
   "source": [
    "# visualizing the first 10 images in the dataset and their labels\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "plt.figure(figsize=(10, 1))\n",
    "for i in range(10):\n",
    "    plt.subplot(1, 10, i+1)\n",
    "    plt.imshow(X_train[i].reshape(32,32),cmap='gray')\n",
    "    plt.axis('off')\n",
    "plt.show()\n",
    "print('label for each of the above image: %s' % (y_train[0:10]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Jag4FxwiUZyh"
   },
   "source": [
    "# 3. Implementing an optimal K-Nearest Neighbour Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "LGNCJDbBUZyj"
   },
   "outputs": [],
   "source": [
    "#Reshape data from 2D to 1D -> 32X32 to 1024\n",
    "X_train = np.asarray(X_train).reshape(42000,1024)\n",
    "X_test = np.asarray(X_test).reshape(18000,1024)\n",
    "X_val = np.asarray(X_val).reshape(60000,1024)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 236
    },
    "colab_type": "code",
    "id": "GeqMgXALl2wp",
    "outputId": "c916fb12-d731-4d2f-e8ee-f31b228ef025"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "k value=1, accuracy score=83.78%\n",
      "k value=3, accuracy score=62.83%\n",
      "k value=5, accuracy score=62.42%\n",
      "k value=7, accuracy score=61.43%\n",
      "k value=9, accuracy score=60.59%\n",
      "k value=11, accuracy score=60.32%\n",
      "k value=13, accuracy score=59.88%\n",
      "k value=15, accuracy score=59.45%\n",
      "k value=17, accuracy score=59.25%\n",
      "k value=19, accuracy score=59.10%\n",
      "k value=21, accuracy score=58.77%\n",
      "k value=23, accuracy score=58.62%\n",
      "k value=25, accuracy score=58.41%\n",
      "k value=27, accuracy score=58.21%\n",
      "k value=29, accuracy score=58.18%\n",
      "k=1 value has highest accuracy of 83.78% on validation data\n"
     ]
    }
   ],
   "source": [
    "# Initializing the value of k and finding the accuracies on validation data\n",
    "k_vals = range(1, 30, 2)\n",
    "accuracies = []\n",
    "\n",
    "for k in range(1, 30, 2):\n",
    "  knn = KNeighborsClassifier(n_neighbors=k)\n",
    "  knn.fit(X_train, y_train)\n",
    "  score = knn.score(X_val, y_val)\n",
    "  print(\"k value=%d, accuracy score=%.2f%%\" % (k, score * 100))\n",
    "  accuracies.append(score)\n",
    " \n",
    "# finding the value of k which has the largest accuracy\n",
    "i = int(np.argmax(accuracies))\n",
    "print(\"k=%d value has highest accuracy of %.2f%% on validation data\" % (k_vals[i],accuracies[i] * 100))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* It took more than 24hrs to find the k values and accuracy of the model using 4GB RAM and intel i5 processor. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "RyVrj74knj1r"
   },
   "outputs": [],
   "source": [
    "# Retraining the model using the best k value and predict the labels on test data\n",
    "\n",
    "knn = KNeighborsClassifier(n_neighbors=k_vals[i])\n",
    "knn.fit(X_train, y_train)\n",
    "predictions = knn.predict(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "oUtEGPE7UZy5"
   },
   "source": [
    " # 4. Printing Classification Metric Report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "eL_HE_HPUZy5",
    "outputId": "408470bd-9488-4ef1-a417-e21a04c46fd3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "EVALUATION ON TESTING DATA\n",
      "[[ 935   43   51   64   71   65  185   40  170  190]\n",
      " [  51 1034  109  122  123   69   68  119   67   66]\n",
      " [  63  135  872  131   82   74   65  168   79  134]\n",
      " [  67  166  127  610   88  229   82   80  149  121]\n",
      " [  85  195   51   71 1039   55  111   35   98   72]\n",
      " [  94  103   71  220   54  588  203   48  205  182]\n",
      " [ 225   74   51   72  102  154  711   33  295  115]\n",
      " [  65  173  146  100   34   38   48 1071   53   80]\n",
      " [ 150   62   59  130   86  153  267   44  675  186]\n",
      " [ 238   87   89  115   60  113  115   65  192  730]]\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.47      0.52      0.49      1814\n",
      "           1       0.50      0.57      0.53      1828\n",
      "           2       0.54      0.48      0.51      1803\n",
      "           3       0.37      0.35      0.36      1719\n",
      "           4       0.60      0.57      0.59      1812\n",
      "           5       0.38      0.33      0.36      1768\n",
      "           6       0.38      0.39      0.39      1832\n",
      "           7       0.63      0.59      0.61      1808\n",
      "           8       0.34      0.37      0.36      1812\n",
      "           9       0.39      0.40      0.40      1804\n",
      "\n",
      "    accuracy                           0.46     18000\n",
      "   macro avg       0.46      0.46      0.46     18000\n",
      "weighted avg       0.46      0.46      0.46     18000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# show a final classification report demonstrating the accuracy of the classifier\n",
    "print(\"EVALUATION ON TESTING DATA\")\n",
    "print(confusion_matrix(y_test,predictions))\n",
    "print(classification_report(y_test, predictions))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The accuracy of the model is 46%.That is not good. The digit 4 and 6 are classified correctly and achieved approx 60% accuracy."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Predicting the digits using KNN Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 144x144 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[9]\n"
     ]
    }
   ],
   "source": [
    "plt.figure(figsize=(2,2))\n",
    "plt.imshow(X_test[59].reshape(32,32))\n",
    "plt.show()\n",
    "image = X_test[15]\n",
    "print(knn.predict(image.reshape(1, -1)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "CKt41c2zUZzO"
   },
   "source": [
    "# 5. Implementing Deep Neural Network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "wRZtrDmiUZzV"
   },
   "outputs": [],
   "source": [
    "#Reshape data from 2D to 1D -> 32X32 to 1024\n",
    "X_train = np.asarray(X_train).reshape(42000,1024)\n",
    "X_test = np.asarray(X_test).reshape(18000,1024)\n",
    "X_val = np.asarray(X_val).reshape(60000,1024)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "rC2jjqr6UZza"
   },
   "source": [
    "#### Converting output label to multiple values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "YoLlsAyFUZzc"
   },
   "outputs": [],
   "source": [
    "y_train = tf.keras.utils.to_categorical(y_train, num_classes=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "VvnWSJ03UZzf"
   },
   "outputs": [],
   "source": [
    "y_test = tf.keras.utils.to_categorical(y_test, num_classes=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "y3rN20T3UZzi"
   },
   "outputs": [],
   "source": [
    "y_val = tf.keras.utils.to_categorical(y_val, num_classes=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "XJUKSgkfUZzr",
    "outputId": "8de9e85a-1ccc-42ab-bb0c-df6510cba1cb",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(42000, 1024) (18000, 1024) (60000, 1024) (42000, 10) (18000, 10) (60000, 10)\n"
     ]
    }
   ],
   "source": [
    "print(X_train.shape, X_test.shape,X_val.shape, y_train.shape, y_test.shape,y_val.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing Neural Network "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 304,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "VhQqcR5nUZzt"
   },
   "outputs": [],
   "source": [
    "##Initialize the Artificial Neural Network Classifier\n",
    "keras_model = Sequential()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 305,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Input Layer\n",
    "#Adding Input layer and activation functions ReLU\n",
    "keras_model.add(Dense(512, kernel_initializer='he_normal',input_shape = (1024, )))\n",
    "#Adding Activation function\n",
    "keras_model.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 1\n",
    "#Adding first Hidden layer\n",
    "keras_model.add(Dense(256, kernel_initializer='he_normal'))\n",
    "#Adding Activation function\n",
    "keras_model.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 2\n",
    "#Adding second Hidden layer\n",
    "keras_model.add(Dense(128, kernel_initializer='he_normal'))\n",
    "#Adding Activation function\n",
    "keras_model.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 3\n",
    "#Adding third Hidden layer\n",
    "keras_model.add(Dense(64, kernel_initializer='he_normal'))\n",
    "#Adding Activation function\n",
    "keras_model.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 4\n",
    "#Adding fourth Hidden layer\n",
    "keras_model.add(Dense(32, kernel_initializer='he_normal'))\n",
    "#Adding Activation function\n",
    "keras_model.add(Activation('relu'))\n",
    "\n",
    "# Output Layer\n",
    "#Adding output layer which is of 10 nodes (digits)\n",
    "keras_model.add(Dense(10))\n",
    "#Adding Activation function\n",
    "# Here, we are using softmax function because we have multiclass classsification\n",
    "keras_model.add(Activation('softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 293,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "J9oI6gOHUZ0I",
    "outputId": "102b2070-724f-4975-9fcf-01719ee31e46"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_1 (Dense)              (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "activation_4 (Activation)    (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "activation_5 (Activation)    (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 10)                330       \n",
      "_________________________________________________________________\n",
      "activation_6 (Activation)    (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 699,690\n",
      "Trainable params: 699,690\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras_model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using sgd optimizer  and Categorical Crossentropy as loss function and Accuracy as the metric to monitor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 247,
   "metadata": {},
   "outputs": [],
   "source": [
    "# compiling the ANN classifier\n",
    "keras_model.compile(optimizer = 'sgd', loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 248,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/10\n",
      "42000/42000 [==============================] - 6s 137us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 2/10\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 3/10\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 4/10\n",
      "42000/42000 [==============================] - 5s 131us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 5/10\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 6/10\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 7/10\n",
      "42000/42000 [==============================] - 6s 133us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 8/10\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 9/10\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 10/10\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history = keras_model.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 10, verbose = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using adam optimizer  and Categorical Crossentropy as loss function and Accuracy as the metric to monitor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 250,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "IoOqs012UZ0L"
   },
   "outputs": [],
   "source": [
    "# compiling the ANN classifier\n",
    "keras_model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 251,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "s_GSON3iUZ0R",
    "outputId": "91870134-3b0b-4ae8-b1ff-9a07ca473091"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/10\n",
      "42000/42000 [==============================] - 7s 175us/step - loss: 2.3027 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 2/10\n",
      "42000/42000 [==============================] - 7s 162us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 3/10\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 4/10\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 2.3027 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 5/10\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 6/10\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 2.3027 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 7/10\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 2.3026 - accuracy: 0.1015 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 8/10\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 2.3026 - accuracy: 0.1006 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 9/10\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n",
      "Epoch 10/10\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 2.3026 - accuracy: 0.1019 - val_loss: 2.3026 - val_accuracy: 0.1000\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history = keras_model.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 10, verbose = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The accuracy of the model on val data is very low either using adam or sgd."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Now, try to change the learning rate in SGD optimizer\n",
    "As, we are taking learning rate small means model will learn slowly. So, we have to take more number of epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 280,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 9.5205 - accuracy: 0.1035 - val_loss: 2.3986 - val_accuracy: 0.1030\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.3534 - accuracy: 0.1020 - val_loss: 2.3296 - val_accuracy: 0.1015\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 5s 123us/step - loss: 2.3226 - accuracy: 0.1006 - val_loss: 2.3164 - val_accuracy: 0.1011\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 5s 120us/step - loss: 2.3139 - accuracy: 0.1007 - val_loss: 2.3110 - val_accuracy: 0.1012\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 5s 120us/step - loss: 2.3098 - accuracy: 0.1009 - val_loss: 2.3083 - val_accuracy: 0.1009\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 5s 122us/step - loss: 2.3075 - accuracy: 0.1009 - val_loss: 2.3066 - val_accuracy: 0.1010\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.3061 - accuracy: 0.1011 - val_loss: 2.3056 - val_accuracy: 0.1010\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 5s 122us/step - loss: 2.3051 - accuracy: 0.1010 - val_loss: 2.3048 - val_accuracy: 0.1011\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 5s 122us/step - loss: 2.3043 - accuracy: 0.1011 - val_loss: 2.3042 - val_accuracy: 0.1011\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 5s 122us/step - loss: 2.3037 - accuracy: 0.1012 - val_loss: 2.3037 - val_accuracy: 0.1012\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.3031 - accuracy: 0.1011 - val_loss: 2.3033 - val_accuracy: 0.1012\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.3026 - accuracy: 0.1015 - val_loss: 2.3029 - val_accuracy: 0.1012\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.3022 - accuracy: 0.1027 - val_loss: 2.3026 - val_accuracy: 0.1013\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.3017 - accuracy: 0.1035 - val_loss: 2.3023 - val_accuracy: 0.1015\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.3014 - accuracy: 0.1034 - val_loss: 2.3020 - val_accuracy: 0.1014\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.3011 - accuracy: 0.1036 - val_loss: 2.3018 - val_accuracy: 0.1014\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.3007 - accuracy: 0.1035 - val_loss: 2.3015 - val_accuracy: 0.1015\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.3004 - accuracy: 0.1037 - val_loss: 2.3013 - val_accuracy: 0.1016\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.3001 - accuracy: 0.1036 - val_loss: 2.3011 - val_accuracy: 0.1015\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2997 - accuracy: 0.1037 - val_loss: 2.3008 - val_accuracy: 0.1016\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2994 - accuracy: 0.1037 - val_loss: 2.3006 - val_accuracy: 0.1016\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2991 - accuracy: 0.1036 - val_loss: 2.3004 - val_accuracy: 0.1016\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2989 - accuracy: 0.1036 - val_loss: 2.3001 - val_accuracy: 0.1017\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2986 - accuracy: 0.1037 - val_loss: 2.3000 - val_accuracy: 0.1016\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2984 - accuracy: 0.1037 - val_loss: 2.2998 - val_accuracy: 0.1019\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2981 - accuracy: 0.1040 - val_loss: 2.2995 - val_accuracy: 0.1019\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2977 - accuracy: 0.1043 - val_loss: 2.2991 - val_accuracy: 0.1021\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2972 - accuracy: 0.1042 - val_loss: 2.2986 - val_accuracy: 0.1023\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.2967 - accuracy: 0.1046 - val_loss: 2.2981 - val_accuracy: 0.1026\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2960 - accuracy: 0.1050 - val_loss: 2.2972 - val_accuracy: 0.1030\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2950 - accuracy: 0.1055 - val_loss: 2.2960 - val_accuracy: 0.1035\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2936 - accuracy: 0.1066 - val_loss: 2.2945 - val_accuracy: 0.1044\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2918 - accuracy: 0.1070 - val_loss: 2.2923 - val_accuracy: 0.1059\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2896 - accuracy: 0.1081 - val_loss: 2.2895 - val_accuracy: 0.1068\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2862 - accuracy: 0.1092 - val_loss: 2.2865 - val_accuracy: 0.1085\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2830 - accuracy: 0.1105 - val_loss: 2.2826 - val_accuracy: 0.1096\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2798 - accuracy: 0.1120 - val_loss: 2.2792 - val_accuracy: 0.1104\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 6s 132us/step - loss: 2.2766 - accuracy: 0.1127 - val_loss: 2.2768 - val_accuracy: 0.1121\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 5s 129us/step - loss: 2.2736 - accuracy: 0.1138 - val_loss: 2.2733 - val_accuracy: 0.1133\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 5s 129us/step - loss: 2.2707 - accuracy: 0.1151 - val_loss: 2.2702 - val_accuracy: 0.1138\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2678 - accuracy: 0.1182 - val_loss: 2.2683 - val_accuracy: 0.1190\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 6s 134us/step - loss: 2.2655 - accuracy: 0.1216 - val_loss: 2.2653 - val_accuracy: 0.1196\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 6s 132us/step - loss: 2.2629 - accuracy: 0.1224 - val_loss: 2.2625 - val_accuracy: 0.1218\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.2606 - accuracy: 0.1234 - val_loss: 2.2601 - val_accuracy: 0.1213\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 6s 132us/step - loss: 2.2583 - accuracy: 0.1234 - val_loss: 2.2579 - val_accuracy: 0.1223\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 5s 129us/step - loss: 2.2562 - accuracy: 0.1240 - val_loss: 2.2572 - val_accuracy: 0.1238\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 6s 133us/step - loss: 2.2543 - accuracy: 0.1244 - val_loss: 2.2551 - val_accuracy: 0.1227\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2523 - accuracy: 0.1250 - val_loss: 2.2541 - val_accuracy: 0.1224\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2506 - accuracy: 0.1251 - val_loss: 2.2508 - val_accuracy: 0.1239\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 5s 129us/step - loss: 2.2492 - accuracy: 0.1249 - val_loss: 2.2499 - val_accuracy: 0.1245\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 5s 131us/step - loss: 2.2476 - accuracy: 0.1255 - val_loss: 2.2478 - val_accuracy: 0.1247\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 5s 131us/step - loss: 2.2457 - accuracy: 0.1260 - val_loss: 2.2464 - val_accuracy: 0.1247 loss: 2.2445 - accura\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 6s 131us/step - loss: 2.2444 - accuracy: 0.1265 - val_loss: 2.2452 - val_accuracy: 0.1250\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 6s 133us/step - loss: 2.2430 - accuracy: 0.1262 - val_loss: 2.2439 - val_accuracy: 0.1248\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2415 - accuracy: 0.1265 - val_loss: 2.2425 - val_accuracy: 0.1259\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2399 - accuracy: 0.1270 - val_loss: 2.2412 - val_accuracy: 0.1258\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2387 - accuracy: 0.1277 - val_loss: 2.2404 - val_accuracy: 0.1265\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 6s 137us/step - loss: 2.2375 - accuracy: 0.1272 - val_loss: 2.2393 - val_accuracy: 0.1265\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 6s 136us/step - loss: 2.2366 - accuracy: 0.1278 - val_loss: 2.2376 - val_accuracy: 0.1262\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 5s 122us/step - loss: 2.2350 - accuracy: 0.1281 - val_loss: 2.2374 - val_accuracy: 0.1262\n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 5s 123us/step - loss: 2.2340 - accuracy: 0.1286 - val_loss: 2.2371 - val_accuracy: 0.1273\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 5s 123us/step - loss: 2.2328 - accuracy: 0.1287 - val_loss: 2.2345 - val_accuracy: 0.1261\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2316 - accuracy: 0.1287 - val_loss: 2.2349 - val_accuracy: 0.1270\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2303 - accuracy: 0.1290 - val_loss: 2.2324 - val_accuracy: 0.1276\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2293 - accuracy: 0.1288 - val_loss: 2.2309 - val_accuracy: 0.1271 loss: - ETA: 0s - l\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2280 - accuracy: 0.1296 - val_loss: 2.2306 - val_accuracy: 0.1298\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2265 - accuracy: 0.1419 - val_loss: 2.2307 - val_accuracy: 0.1418\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2258 - accuracy: 0.1456 - val_loss: 2.2276 - val_accuracy: 0.1442\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2246 - accuracy: 0.1459 - val_loss: 2.2274 - val_accuracy: 0.1443\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 5s 123us/step - loss: 2.2235 - accuracy: 0.1467 - val_loss: 2.2260 - val_accuracy: 0.1446\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2224 - accuracy: 0.1470 - val_loss: 2.2250 - val_accuracy: 0.1451\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2212 - accuracy: 0.1475 - val_loss: 2.2235 - val_accuracy: 0.1456\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.2204 - accuracy: 0.1484 - val_loss: 2.2230 - val_accuracy: 0.1454\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.2193 - accuracy: 0.1484 - val_loss: 2.2227 - val_accuracy: 0.1452\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2184 - accuracy: 0.1486 - val_loss: 2.2221 - val_accuracy: 0.1474\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 6s 133us/step - loss: 2.2176 - accuracy: 0.1490 - val_loss: 2.2201 - val_accuracy: 0.1469\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 6s 135us/step - loss: 2.2166 - accuracy: 0.1495 - val_loss: 2.2198 - val_accuracy: 0.1480\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 5s 129us/step - loss: 2.2159 - accuracy: 0.1500 - val_loss: 2.2196 - val_accuracy: 0.1475\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 6s 133us/step - loss: 2.2146 - accuracy: 0.1503 - val_loss: 2.2186 - val_accuracy: 0.1471\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2141 - accuracy: 0.1505 - val_loss: 2.2175 - val_accuracy: 0.1488\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2129 - accuracy: 0.1507 - val_loss: 2.2165 - val_accuracy: 0.1492\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 5s 123us/step - loss: 2.2119 - accuracy: 0.1515 - val_loss: 2.2157 - val_accuracy: 0.1495\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2111 - accuracy: 0.1515 - val_loss: 2.2145 - val_accuracy: 0.1499\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2102 - accuracy: 0.1515 - val_loss: 2.2158 - val_accuracy: 0.1479\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2096 - accuracy: 0.1520 - val_loss: 2.2130 - val_accuracy: 0.1495\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2085 - accuracy: 0.1523 - val_loss: 2.2121 - val_accuracy: 0.1510\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2078 - accuracy: 0.1529 - val_loss: 2.2108 - val_accuracy: 0.1506\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2069 - accuracy: 0.1527 - val_loss: 2.2109 - val_accuracy: 0.1506\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.2062 - accuracy: 0.1530 - val_loss: 2.2093 - val_accuracy: 0.1514\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 5s 127us/step - loss: 2.2048 - accuracy: 0.1532 - val_loss: 2.2095 - val_accuracy: 0.1514\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 5s 124us/step - loss: 2.2043 - accuracy: 0.1533 - val_loss: 2.2124 - val_accuracy: 0.1522\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 5s 130us/step - loss: 2.2034 - accuracy: 0.1542 - val_loss: 2.2089 - val_accuracy: 0.1516\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 5s 128us/step - loss: 2.2028 - accuracy: 0.1546 - val_loss: 2.2060 - val_accuracy: 0.1531\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 6s 135us/step - loss: 2.2018 - accuracy: 0.1546 - val_loss: 2.2057 - val_accuracy: 0.1531\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 5s 126us/step - loss: 2.2014 - accuracy: 0.1550 - val_loss: 2.2049 - val_accuracy: 0.1535\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 6s 137us/step - loss: 2.2001 - accuracy: 0.1555 - val_loss: 2.2038 - val_accuracy: 0.153515\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 5s 111us/step - loss: 2.1996 - accuracy: 0.1555 - val_loss: 2.2032 - val_accuracy: 0.1542\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 5s 121us/step - loss: 2.1985 - accuracy: 0.1562 - val_loss: 2.2020 - val_accuracy: 0.1542\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.1978 - accuracy: 0.1568 - val_loss: 2.2014 - val_accuracy: 0.1548\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 5s 125us/step - loss: 2.1969 - accuracy: 0.1568 - val_loss: 2.2017 - val_accuracy: 0.1553\n"
     ]
    }
   ],
   "source": [
    "#Setting the learning rate  and decay in sgd\n",
    "sgd = optimizers.sgd(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model.compile(optimizer = sgd, loss = 'categorical_crossentropy', metrics = ['accuracy'])\n",
    "# Fitting the ANN to the Training data\n",
    "history = keras_model.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The accuracy of the model on val data using sgd with learning rate=0.0001 is improving a bit but not much."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Now, let's try to change the learning rate in ADAM optimizer\n",
    "As, we are taking learning rate small means model will learn slowly. So, we have to take more number of epochs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 306,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 8s 180us/step - loss: 10.9612 - accuracy: 0.1337 - val_loss: 4.1578 - val_accuracy: 0.1754\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 3.0128 - accuracy: 0.1697 - val_loss: 2.3374 - val_accuracy: 0.1452\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 2.2645 - accuracy: 0.1620 - val_loss: 2.2029 - val_accuracy: 0.1779\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 2.1837 - accuracy: 0.1856 - val_loss: 2.1445 - val_accuracy: 0.1997\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 2.1190 - accuracy: 0.2090 - val_loss: 2.0763 - val_accuracy: 0.2268\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 2.0658 - accuracy: 0.2303 - val_loss: 2.0505 - val_accuracy: 0.2351\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 2.0183 - accuracy: 0.2509 - val_loss: 1.9865 - val_accuracy: 0.2644\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 7s 160us/step - loss: 1.9855 - accuracy: 0.2660 - val_loss: 1.9963 - val_accuracy: 0.2831\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 7s 163us/step - loss: 1.9596 - accuracy: 0.2761 - val_loss: 1.9294 - val_accuracy: 0.2909\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 7s 163us/step - loss: 1.9308 - accuracy: 0.2870 - val_loss: 1.8983 - val_accuracy: 0.2988\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 7s 163us/step - loss: 1.8966 - accuracy: 0.3053 - val_loss: 1.8675 - val_accuracy: 0.3195\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 7s 161us/step - loss: 1.8727 - accuracy: 0.3160 - val_loss: 1.8467 - val_accuracy: 0.3271\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 1.8341 - accuracy: 0.3287 - val_loss: 1.8670 - val_accuracy: 0.3146\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.8194 - accuracy: 0.3368 - val_loss: 1.8491 - val_accuracy: 0.3309\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 1.7954 - accuracy: 0.3472 - val_loss: 1.8495 - val_accuracy: 0.3487\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 7s 171us/step - loss: 1.7754 - accuracy: 0.3570 - val_loss: 1.7312 - val_accuracy: 0.3736\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 1.7514 - accuracy: 0.3721 - val_loss: 1.7750 - val_accuracy: 0.3501\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.7117 - accuracy: 0.3886 - val_loss: 1.7157 - val_accuracy: 0.4004\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.7069 - accuracy: 0.3933 - val_loss: 1.6392 - val_accuracy: 0.4139\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 7s 162us/step - loss: 1.6656 - accuracy: 0.4061 - val_loss: 1.6398 - val_accuracy: 0.4358\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 7s 171us/step - loss: 1.6203 - accuracy: 0.4277 - val_loss: 1.6274 - val_accuracy: 0.4196\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 1.6323 - accuracy: 0.4245 - val_loss: 1.6161 - val_accuracy: 0.4223\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.5864 - accuracy: 0.4457 - val_loss: 1.6242 - val_accuracy: 0.4434\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.5896 - accuracy: 0.4445 - val_loss: 1.6270 - val_accuracy: 0.4251\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.5546 - accuracy: 0.4595 - val_loss: 1.5303 - val_accuracy: 0.4783\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.5173 - accuracy: 0.4765 - val_loss: 1.5563 - val_accuracy: 0.4958\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.4807 - accuracy: 0.4911 - val_loss: 1.4779 - val_accuracy: 0.5108\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 1.4540 - accuracy: 0.4978 - val_loss: 1.4808 - val_accuracy: 0.4870\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 1.4127 - accuracy: 0.5144 - val_loss: 1.4253 - val_accuracy: 0.5143\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 1.3928 - accuracy: 0.5232 - val_loss: 1.4046 - val_accuracy: 0.5365\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 1.3814 - accuracy: 0.5261 - val_loss: 1.3825 - val_accuracy: 0.5419\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 1.3503 - accuracy: 0.5388 - val_loss: 1.3604 - val_accuracy: 0.5342\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.3387 - accuracy: 0.5408 - val_loss: 1.3600 - val_accuracy: 0.5438\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.3211 - accuracy: 0.5477 - val_loss: 1.3480 - val_accuracy: 0.5480\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 1.3189 - accuracy: 0.5498 - val_loss: 1.3754 - val_accuracy: 0.5544\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 7s 172us/step - loss: 1.2863 - accuracy: 0.5585 - val_loss: 1.3406 - val_accuracy: 0.5303\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.2841 - accuracy: 0.5605 - val_loss: 1.2671 - val_accuracy: 0.5769\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 1.2399 - accuracy: 0.5730 - val_loss: 1.2670 - val_accuracy: 0.5688\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.2417 - accuracy: 0.5717 - val_loss: 1.2635 - val_accuracy: 0.5666\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 1.1905 - accuracy: 0.5882 - val_loss: 1.2216 - val_accuracy: 0.5898\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 1.1886 - accuracy: 0.5862 - val_loss: 1.2398 - val_accuracy: 0.5793\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 1.1591 - accuracy: 0.6001 - val_loss: 1.2095 - val_accuracy: 0.5874\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 8s 188us/step - loss: 1.1536 - accuracy: 0.5996 - val_loss: 1.2058 - val_accuracy: 0.5830\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 8s 181us/step - loss: 1.1317 - accuracy: 0.6100 - val_loss: 1.1766 - val_accuracy: 0.5973\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 8s 186us/step - loss: 1.1185 - accuracy: 0.6132 - val_loss: 1.1847 - val_accuracy: 0.6016\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 7s 173us/step - loss: 1.0920 - accuracy: 0.6270 - val_loss: 1.0943 - val_accuracy: 0.6280\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 1.0811 - accuracy: 0.6334 - val_loss: 1.1102 - val_accuracy: 0.6280\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 1.0426 - accuracy: 0.6564 - val_loss: 1.0889 - val_accuracy: 0.6398\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 1.0240 - accuracy: 0.6617 - val_loss: 1.0394 - val_accuracy: 0.6637\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 1.0089 - accuracy: 0.6699 - val_loss: 1.0904 - val_accuracy: 0.6456\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 8s 179us/step - loss: 0.9735 - accuracy: 0.6805 - val_loss: 1.0256 - val_accuracy: 0.6772\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.9530 - accuracy: 0.6910 - val_loss: 0.9642 - val_accuracy: 0.6980\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 7s 164us/step - loss: 0.9383 - accuracy: 0.6957 - val_loss: 0.9404 - val_accuracy: 0.7048\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 7s 171us/step - loss: 0.9179 - accuracy: 0.7033 - val_loss: 1.0039 - val_accuracy: 0.6853\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 7s 172us/step - loss: 0.9056 - accuracy: 0.7099 - val_loss: 0.9183 - val_accuracy: 0.7122\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 0.8802 - accuracy: 0.7199 - val_loss: 0.9217 - val_accuracy: 0.7146\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.8704 - accuracy: 0.7205 - val_loss: 0.9043 - val_accuracy: 0.7172\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.8315 - accuracy: 0.7349 - val_loss: 0.8674 - val_accuracy: 0.7322\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.8276 - accuracy: 0.7359 - val_loss: 0.8790 - val_accuracy: 0.7318\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 7s 173us/step - loss: 0.8156 - accuracy: 0.7390 - val_loss: 0.8474 - val_accuracy: 0.74375 \n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 7s 172us/step - loss: 0.7861 - accuracy: 0.7490 - val_loss: 0.8758 - val_accuracy: 0.7285\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 8s 187us/step - loss: 0.7745 - accuracy: 0.7557 - val_loss: 0.8352 - val_accuracy: 0.7466\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 8s 197us/step - loss: 0.7653 - accuracy: 0.7569 - val_loss: 0.8503 - val_accuracy: 0.7389\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 7s 171us/step - loss: 0.7646 - accuracy: 0.7581 - val_loss: 0.8016 - val_accuracy: 0.7572\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 7s 175us/step - loss: 0.7403 - accuracy: 0.7654 - val_loss: 0.7914 - val_accuracy: 0.7561\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.7321 - accuracy: 0.7685 - val_loss: 0.8259 - val_accuracy: 0.7490\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.7063 - accuracy: 0.7762 - val_loss: 0.7652 - val_accuracy: 0.7691\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.7154 - accuracy: 0.7735 - val_loss: 0.7766 - val_accuracy: 0.7649\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 0.6961 - accuracy: 0.7810 - val_loss: 0.8603 - val_accuracy: 0.7427\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.7065 - accuracy: 0.7756 - val_loss: 0.7914 - val_accuracy: 0.7600\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.6784 - accuracy: 0.7857 - val_loss: 0.8381 - val_accuracy: 0.7490\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.6846 - accuracy: 0.7830 - val_loss: 0.7265 - val_accuracy: 0.7794\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.6523 - accuracy: 0.7951 - val_loss: 0.7404 - val_accuracy: 0.7802\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 7s 172us/step - loss: 0.6497 - accuracy: 0.7954 - val_loss: 0.7170 - val_accuracy: 0.7861\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 7s 173us/step - loss: 0.6417 - accuracy: 0.7971 - val_loss: 0.6997 - val_accuracy: 0.7914\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.6326 - accuracy: 0.7987 - val_loss: 0.7423 - val_accuracy: 0.7719\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 7s 174us/step - loss: 0.6251 - accuracy: 0.8023 - val_loss: 0.7154 - val_accuracy: 0.7900\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 7s 178us/step - loss: 0.6165 - accuracy: 0.8050 - val_loss: 0.7085 - val_accuracy: 0.7883\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 0.6139 - accuracy: 0.8060 - val_loss: 0.6747 - val_accuracy: 0.7997\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 0.5947 - accuracy: 0.8126 - val_loss: 0.7379 - val_accuracy: 0.7823\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.5988 - accuracy: 0.8099 - val_loss: 0.6962 - val_accuracy: 0.7946\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.5844 - accuracy: 0.8160 - val_loss: 0.6771 - val_accuracy: 0.7985\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 0.5851 - accuracy: 0.8128 - val_loss: 0.7049 - val_accuracy: 0.7890\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.5728 - accuracy: 0.8201 - val_loss: 0.6947 - val_accuracy: 0.7936\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.5697 - accuracy: 0.8206 - val_loss: 0.6805 - val_accuracy: 0.7935\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.5682 - accuracy: 0.8187 - val_loss: 0.6217 - val_accuracy: 0.8164\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.5519 - accuracy: 0.8255 - val_loss: 0.6732 - val_accuracy: 0.8043.5\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 7s 171us/step - loss: 0.5448 - accuracy: 0.8291 - val_loss: 0.6394 - val_accuracy: 0.8109\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 7s 167us/step - loss: 0.5393 - accuracy: 0.8279 - val_loss: 0.6069 - val_accuracy: 0.8238\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.5247 - accuracy: 0.8352 - val_loss: 0.6250 - val_accuracy: 0.8140\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.5331 - accuracy: 0.8289 - val_loss: 0.5951 - val_accuracy: 0.8260\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.5147 - accuracy: 0.8358 - val_loss: 0.6245 - val_accuracy: 0.8184\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 0.5198 - accuracy: 0.8342 - val_loss: 0.6407 - val_accuracy: 0.8171\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.5091 - accuracy: 0.8372 - val_loss: 0.6105 - val_accuracy: 0.8233\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.5115 - accuracy: 0.8387 - val_loss: 0.6595 - val_accuracy: 0.8058\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 7s 166us/step - loss: 0.5085 - accuracy: 0.8388 - val_loss: 0.5948 - val_accuracy: 0.8266\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.4869 - accuracy: 0.8450 - val_loss: 0.5993 - val_accuracy: 0.8271\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 7s 170us/step - loss: 0.4812 - accuracy: 0.8459 - val_loss: 0.5610 - val_accuracy: 0.8411\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.4714 - accuracy: 0.8515 - val_loss: 0.6215 - val_accuracy: 0.8184\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.4757 - accuracy: 0.8493 - val_loss: 0.6009 - val_accuracy: 0.8264\n"
     ]
    }
   ],
   "source": [
    "#Setting the learning rate  and decay in sgd\n",
    "adam = optimizers.adam(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model.compile(optimizer = adam, loss = 'categorical_crossentropy', metrics = ['accuracy'])\n",
    "# Fitting the ANN to the Training data\n",
    "history = keras_model.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The accuracy of the model on val data using ADAM with learning rate=0.0001 has shown significant improvement."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 307,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "BviBAo04UZ0W"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60000/60000 [==============================] - 5s 87us/step\n",
      "Val_acc using simple NN adam :  0.8263999819755554\n"
     ]
    }
   ],
   "source": [
    "results = keras_model.evaluate(X_val, y_val)\n",
    "print('Val_acc using simple NN adam : ', results[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 308,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Method</th>\n",
       "      <th>accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>NN</td>\n",
       "      <td>0.8264</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  Method  accuracy\n",
       "1     NN    0.8264"
      ]
     },
     "execution_count": 308,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Store the accuracy results for each model in a dataframe for final comparison\n",
    "results_on_val = pd.DataFrame({'Method':['NN'], 'accuracy': results[1]},index={'1'})\n",
    "results_on_val = results_on_val[['Method', 'accuracy']]\n",
    "results_on_val"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing batch normalization for training the neural network "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 309,
   "metadata": {},
   "outputs": [],
   "source": [
    "##Initialize the Artificial Neural Network Classifier\n",
    "keras_model_1 = Sequential()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Adding Batch normalization layer "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 310,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Input Layer\n",
    "#Adding Input layer and activation functions ReLU\n",
    "keras_model_1.add(Dense(512, kernel_initializer='he_normal',input_shape = (1024, )))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_1.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_1.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 1\n",
    "#Adding first Hidden layer\n",
    "keras_model_1.add(Dense(256, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_1.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_1.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 2\n",
    "#Adding second Hidden layer\n",
    "keras_model_1.add(Dense(128, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_1.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_1.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 3\n",
    "#Adding third Hidden layer\n",
    "keras_model_1.add(Dense(64, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_1.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_1.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 4\n",
    "#Adding fourth Hidden layer\n",
    "keras_model_1.add(Dense(32, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_1.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_1.add(Activation('relu'))\n",
    "\n",
    "# Output Layer\n",
    "#Adding output layer which is of 10 nodes (digidts)\n",
    "keras_model_1.add(Dense(10))\n",
    "#Adding Activation function\n",
    "# Here, we are using softmax function because we have multiclass classsification\n",
    "keras_model_1.add(Activation('softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 311,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_3\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_13 (Dense)             (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "activation_13 (Activation)   (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_14 (Dense)             (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 256)               1024      \n",
      "_________________________________________________________________\n",
      "activation_14 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_15 (Dense)             (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_15 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_16 (Dense)             (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "batch_normalization_4 (Batch (None, 64)                256       \n",
      "_________________________________________________________________\n",
      "activation_16 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_17 (Dense)             (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 32)                128       \n",
      "_________________________________________________________________\n",
      "activation_17 (Activation)   (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_18 (Dense)             (None, 10)                330       \n",
      "_________________________________________________________________\n",
      "activation_18 (Activation)   (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 703,658\n",
      "Trainable params: 701,674\n",
      "Non-trainable params: 1,984\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras_model_1.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using ADAM optimizer.\n",
    "* learning rate= 0.0001\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 312,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Setting the learning rate  and decay in sgd\n",
    "adam = optimizers.adam(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model_1.compile(optimizer = adam, loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 313,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 10s 248us/step - loss: 2.1479 - accuracy: 0.2508 - val_loss: 1.9940 - val_accuracy: 0.3037\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 10s 228us/step - loss: 1.6269 - accuracy: 0.4861 - val_loss: 1.6468 - val_accuracy: 0.4754\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 9s 208us/step - loss: 1.3413 - accuracy: 0.6087 - val_loss: 1.4551 - val_accuracy: 0.5534\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 1.1435 - accuracy: 0.6808 - val_loss: 1.2232 - val_accuracy: 0.6539\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.9963 - accuracy: 0.7241 - val_loss: 1.1230 - val_accuracy: 0.6831\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.8880 - accuracy: 0.7514 - val_loss: 1.0722 - val_accuracy: 0.6806\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.7987 - accuracy: 0.7750 - val_loss: 1.0617 - val_accuracy: 0.6705\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.7276 - accuracy: 0.7931 - val_loss: 0.9050 - val_accuracy: 0.7326\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.6695 - accuracy: 0.8090 - val_loss: 0.9625 - val_accuracy: 0.7067\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.6169 - accuracy: 0.8217 - val_loss: 0.8344 - val_accuracy: 0.7462\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.5752 - accuracy: 0.8326 - val_loss: 0.8745 - val_accuracy: 0.7266\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.5344 - accuracy: 0.8450 - val_loss: 0.8295 - val_accuracy: 0.7479\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.4986 - accuracy: 0.8542 - val_loss: 0.7200 - val_accuracy: 0.7804\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.4664 - accuracy: 0.8652 - val_loss: 0.7070 - val_accuracy: 0.7868\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.4434 - accuracy: 0.8694 - val_loss: 0.7443 - val_accuracy: 0.7682\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.4158 - accuracy: 0.8783 - val_loss: 0.7295 - val_accuracy: 0.7746\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 9s 222us/step - loss: 0.3950 - accuracy: 0.8838 - val_loss: 0.7645 - val_accuracy: 0.7680\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.3766 - accuracy: 0.8886 - val_loss: 0.6388 - val_accuracy: 0.8005\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.3513 - accuracy: 0.8969 - val_loss: 0.6649 - val_accuracy: 0.7917\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.3337 - accuracy: 0.9038 - val_loss: 0.7547 - val_accuracy: 0.7712\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.3151 - accuracy: 0.9080 - val_loss: 0.6774 - val_accuracy: 0.7919\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.2967 - accuracy: 0.9154 - val_loss: 0.6267 - val_accuracy: 0.8054\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.2844 - accuracy: 0.9178 - val_loss: 0.7990 - val_accuracy: 0.7590\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.2703 - accuracy: 0.9220 - val_loss: 0.6548 - val_accuracy: 0.8005\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 9s 224us/step - loss: 0.2582 - accuracy: 0.9265 - val_loss: 0.6855 - val_accuracy: 0.7929\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.2434 - accuracy: 0.9301 - val_loss: 0.6434 - val_accuracy: 0.8034\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.2374 - accuracy: 0.9308 - val_loss: 0.8516 - val_accuracy: 0.7497\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.2257 - accuracy: 0.9353 - val_loss: 0.6059 - val_accuracy: 0.8120\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.2128 - accuracy: 0.9386 - val_loss: 0.7449 - val_accuracy: 0.7832\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.2089 - accuracy: 0.9390 - val_loss: 0.5357 - val_accuracy: 0.8359\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1981 - accuracy: 0.9427 - val_loss: 0.6945 - val_accuracy: 0.7930\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.1866 - accuracy: 0.9469 - val_loss: 0.8193 - val_accuracy: 0.7649\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.1799 - accuracy: 0.9492 - val_loss: 0.7204 - val_accuracy: 0.7848\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.1799 - accuracy: 0.9487 - val_loss: 0.6678 - val_accuracy: 0.8050\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.1709 - accuracy: 0.9509 - val_loss: 0.6431 - val_accuracy: 0.8050\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.1587 - accuracy: 0.9558 - val_loss: 0.5821 - val_accuracy: 0.8269\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.1572 - accuracy: 0.9554 - val_loss: 0.7113 - val_accuracy: 0.7919\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.1564 - accuracy: 0.9546 - val_loss: 0.5533 - val_accuracy: 0.8348\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.1471 - accuracy: 0.9577 - val_loss: 0.6933 - val_accuracy: 0.7944\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.1424 - accuracy: 0.9580 - val_loss: 0.6980 - val_accuracy: 0.7987\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.1307 - accuracy: 0.9631 - val_loss: 0.7006 - val_accuracy: 0.8103\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.1309 - accuracy: 0.9636 - val_loss: 0.7381 - val_accuracy: 0.7944\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1239 - accuracy: 0.9646 - val_loss: 0.6867 - val_accuracy: 0.8049\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.1194 - accuracy: 0.9664 - val_loss: 0.6597 - val_accuracy: 0.8183\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.1173 - accuracy: 0.9666 - val_loss: 0.5662 - val_accuracy: 0.8384\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.1140 - accuracy: 0.9685 - val_loss: 0.6105 - val_accuracy: 0.8279\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.1105 - accuracy: 0.9691 - val_loss: 0.5682 - val_accuracy: 0.8372\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.1132 - accuracy: 0.9681 - val_loss: 0.6688 - val_accuracy: 0.8207\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.1040 - accuracy: 0.9698 - val_loss: 0.5072 - val_accuracy: 0.8566\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0966 - accuracy: 0.9732 - val_loss: 0.5099 - val_accuracy: 0.8571\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 9s 219us/step - loss: 0.1029 - accuracy: 0.9705 - val_loss: 0.8159 - val_accuracy: 0.7912\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 9s 219us/step - loss: 0.1000 - accuracy: 0.9717 - val_loss: 0.7615 - val_accuracy: 0.8057\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.0910 - accuracy: 0.9740 - val_loss: 0.6439 - val_accuracy: 0.8251\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0864 - accuracy: 0.9756 - val_loss: 0.8333 - val_accuracy: 0.7959\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.0915 - accuracy: 0.9731 - val_loss: 0.6697 - val_accuracy: 0.8232\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 10s 229us/step - loss: 0.0920 - accuracy: 0.9730 - val_loss: 0.6388 - val_accuracy: 0.8267\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0906 - accuracy: 0.9732 - val_loss: 0.6592 - val_accuracy: 0.8146\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0825 - accuracy: 0.9770 - val_loss: 0.5564 - val_accuracy: 0.8516\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0800 - accuracy: 0.9768 - val_loss: 0.6041 - val_accuracy: 0.8368\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0763 - accuracy: 0.9777 - val_loss: 0.7079 - val_accuracy: 0.8241\n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 9s 205us/step - loss: 0.0706 - accuracy: 0.9800 - val_loss: 0.5665 - val_accuracy: 0.8557\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 8s 195us/step - loss: 0.0700 - accuracy: 0.9809 - val_loss: 0.5185 - val_accuracy: 0.8595\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0813 - accuracy: 0.9765 - val_loss: 0.8970 - val_accuracy: 0.7693\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 11s 268us/step - loss: 0.0781 - accuracy: 0.9769 - val_loss: 0.6840 - val_accuracy: 0.8230\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 11s 261us/step - loss: 0.0743 - accuracy: 0.9787 - val_loss: 0.6976 - val_accuracy: 0.8199\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 8s 198us/step - loss: 0.0758 - accuracy: 0.9779 - val_loss: 0.6572 - val_accuracy: 0.8311\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 9s 222us/step - loss: 0.0680 - accuracy: 0.9800 - val_loss: 0.6265 - val_accuracy: 0.8433\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 12s 278us/step - loss: 0.0655 - accuracy: 0.9815 - val_loss: 0.5756 - val_accuracy: 0.8501\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 0.0648 - accuracy: 0.9808 - val_loss: 0.6767 - val_accuracy: 0.8245\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 9s 222us/step - loss: 0.0663 - accuracy: 0.9805 - val_loss: 0.6002 - val_accuracy: 0.8470\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 0.0649 - accuracy: 0.9808 - val_loss: 0.6506 - val_accuracy: 0.8266\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0638 - accuracy: 0.9817 - val_loss: 0.6289 - val_accuracy: 0.8425\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0577 - accuracy: 0.9839 - val_loss: 0.5725 - val_accuracy: 0.8575\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 9s 207us/step - loss: 0.0599 - accuracy: 0.9827 - val_loss: 0.8238 - val_accuracy: 0.8018\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0580 - accuracy: 0.9831 - val_loss: 0.7341 - val_accuracy: 0.8230\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 9s 207us/step - loss: 0.0602 - accuracy: 0.9819 - val_loss: 1.0562 - val_accuracy: 0.7669\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0644 - accuracy: 0.9802 - val_loss: 0.8602 - val_accuracy: 0.7997\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0617 - accuracy: 0.9817 - val_loss: 0.7482 - val_accuracy: 0.8302\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.0608 - accuracy: 0.9822 - val_loss: 0.6637 - val_accuracy: 0.8371\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0551 - accuracy: 0.9839 - val_loss: 0.6200 - val_accuracy: 0.8451\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.0541 - accuracy: 0.9847 - val_loss: 0.7855 - val_accuracy: 0.8245\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0510 - accuracy: 0.9855 - val_loss: 0.7493 - val_accuracy: 0.8253\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0522 - accuracy: 0.9850 - val_loss: 0.6144 - val_accuracy: 0.8467\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0534 - accuracy: 0.9845 - val_loss: 0.6563 - val_accuracy: 0.8360\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0466 - accuracy: 0.9862 - val_loss: 0.8250 - val_accuracy: 0.8027\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 9s 210us/step - loss: 0.0545 - accuracy: 0.9838 - val_loss: 0.7299 - val_accuracy: 0.8288\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0504 - accuracy: 0.9851 - val_loss: 0.6813 - val_accuracy: 0.8321\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 9s 208us/step - loss: 0.0517 - accuracy: 0.9851 - val_loss: 0.6487 - val_accuracy: 0.8444\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0456 - accuracy: 0.9864 - val_loss: 0.5678 - val_accuracy: 0.8657\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 8s 192us/step - loss: 0.0450 - accuracy: 0.9871 - val_loss: 0.6896 - val_accuracy: 0.8372\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 10s 229us/step - loss: 0.0541 - accuracy: 0.9835 - val_loss: 0.9403 - val_accuracy: 0.7973\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 9s 222us/step - loss: 0.0470 - accuracy: 0.9864 - val_loss: 0.7333 - val_accuracy: 0.8234\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0517 - accuracy: 0.9848 - val_loss: 0.7947 - val_accuracy: 0.8192\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0423 - accuracy: 0.9874 - val_loss: 0.7737 - val_accuracy: 0.8161\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0433 - accuracy: 0.9875 - val_loss: 0.7116 - val_accuracy: 0.8338\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0468 - accuracy: 0.9862 - val_loss: 0.8658 - val_accuracy: 0.8037\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0475 - accuracy: 0.9850 - val_loss: 0.6512 - val_accuracy: 0.8514\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0431 - accuracy: 0.9870 - val_loss: 0.6795 - val_accuracy: 0.8482\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.0416 - accuracy: 0.9877 - val_loss: 0.7038 - val_accuracy: 0.8354\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0454 - accuracy: 0.9865 - val_loss: 0.7568 - val_accuracy: 0.8315\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history_1 = keras_model_1.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 314,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60000/60000 [==============================] - 6s 107us/step\n",
      "Val_acc using Batch ANN adam :  0.8314666748046875\n"
     ]
    }
   ],
   "source": [
    "results_1 = keras_model_1.evaluate(X_val, y_val)\n",
    "print('Val_acc using Batch ANN adam : ', results_1[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 315,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Method</th>\n",
       "      <th>accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>NN</td>\n",
       "      <td>0.826400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>NN_Batch_adam</td>\n",
       "      <td>0.831467</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Method  accuracy\n",
       "1             NN  0.826400\n",
       "2  NN_Batch_adam  0.831467"
      ]
     },
     "execution_count": 315,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Store the accuracy results for each model in a dataframe for final comparison\n",
    "tempResultsDf = pd.DataFrame({'Method':['NN_Batch_adam'], 'accuracy': [results_1[1]]},index={'2'})\n",
    "results_on_val= pd.concat([results_on_val, tempResultsDf])\n",
    "results_on_val = results_on_val[['Method', 'accuracy']]\n",
    "results_on_val"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing (vectorized) backpropagation (cost stochastic gradient descent, cross entropy loss, cost functions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 316,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Initialize the Artificial Neural Network Classifier\n",
    "keras_model_2 = Sequential()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Adding Batch normalization layer  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 317,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Input Layer\n",
    "#Adding Input layer and activation functions ReLU\n",
    "keras_model_2.add(Dense(512, kernel_initializer='he_normal',input_shape = (1024, )))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_2.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_2.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 1\n",
    "#Adding first Hidden layer\n",
    "keras_model_2.add(Dense(256, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_2.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_2.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 2\n",
    "#Adding second Hidden layer\n",
    "keras_model_2.add(Dense(128, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_2.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_2.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 3\n",
    "#Adding third Hidden layer\n",
    "keras_model_2.add(Dense(64, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_2.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_2.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 4\n",
    "#Adding fourth Hidden layer\n",
    "keras_model_2.add(Dense(32, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_2.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_2.add(Activation('relu'))\n",
    "\n",
    "# Output Layer\n",
    "#Adding output layer which is of 10 nodes (digidts)\n",
    "keras_model_2.add(Dense(10))\n",
    "#Adding Activation function\n",
    "# Here, we are using softmax function because we have multiclass classsification\n",
    "keras_model_2.add(Activation('softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 318,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_4\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_19 (Dense)             (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "batch_normalization_6 (Batch (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "activation_19 (Activation)   (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_20 (Dense)             (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "batch_normalization_7 (Batch (None, 256)               1024      \n",
      "_________________________________________________________________\n",
      "activation_20 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_21 (Dense)             (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "batch_normalization_8 (Batch (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_21 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_22 (Dense)             (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "batch_normalization_9 (Batch (None, 64)                256       \n",
      "_________________________________________________________________\n",
      "activation_22 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_23 (Dense)             (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "batch_normalization_10 (Batc (None, 32)                128       \n",
      "_________________________________________________________________\n",
      "activation_23 (Activation)   (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_24 (Dense)             (None, 10)                330       \n",
      "_________________________________________________________________\n",
      "activation_24 (Activation)   (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 703,658\n",
      "Trainable params: 701,674\n",
      "Non-trainable params: 1,984\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras_model_2.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using SGD optimizer\n",
    "* learning rate =0.0001\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 319,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Setting the learning rate in sgd\n",
    "sgd = optimizers.adam(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model_2.compile(optimizer = sgd, loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 320,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 10s 250us/step - loss: 2.1492 - accuracy: 0.2520 - val_loss: 2.0291 - val_accuracy: 0.2933\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 1.6157 - accuracy: 0.5134 - val_loss: 1.6216 - val_accuracy: 0.5163\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 1.3135 - accuracy: 0.6361 - val_loss: 1.3621 - val_accuracy: 0.6015\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 1.1190 - accuracy: 0.6955 - val_loss: 1.2618 - val_accuracy: 0.6380\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.9821 - accuracy: 0.7283 - val_loss: 1.0875 - val_accuracy: 0.6824\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.8780 - accuracy: 0.7568 - val_loss: 1.0932 - val_accuracy: 0.6722\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 9s 219us/step - loss: 0.7922 - accuracy: 0.7775 - val_loss: 1.0416 - val_accuracy: 0.6826\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 10s 231us/step - loss: 0.7219 - accuracy: 0.7955 - val_loss: 0.9094 - val_accuracy: 0.7325\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.6600 - accuracy: 0.8109 - val_loss: 0.8632 - val_accuracy: 0.7451\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.6138 - accuracy: 0.8243 - val_loss: 0.8541 - val_accuracy: 0.7404\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.5642 - accuracy: 0.8389 - val_loss: 0.8692 - val_accuracy: 0.7369\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.5279 - accuracy: 0.8489 - val_loss: 0.8304 - val_accuracy: 0.7454\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.4982 - accuracy: 0.8565 - val_loss: 0.7686 - val_accuracy: 0.7711\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.4694 - accuracy: 0.8637 - val_loss: 0.7889 - val_accuracy: 0.7487\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.4344 - accuracy: 0.8748 - val_loss: 0.7578 - val_accuracy: 0.7656\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.4154 - accuracy: 0.8787 - val_loss: 0.6822 - val_accuracy: 0.7912\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.3848 - accuracy: 0.8890 - val_loss: 0.9365 - val_accuracy: 0.7234\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.3650 - accuracy: 0.8940 - val_loss: 0.6697 - val_accuracy: 0.7938\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.3461 - accuracy: 0.8998 - val_loss: 0.6536 - val_accuracy: 0.7954\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.3274 - accuracy: 0.9052 - val_loss: 0.5905 - val_accuracy: 0.8183\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.3107 - accuracy: 0.9114 - val_loss: 0.7531 - val_accuracy: 0.7679\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 9s 219us/step - loss: 0.2968 - accuracy: 0.9136 - val_loss: 0.7007 - val_accuracy: 0.7865\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.2837 - accuracy: 0.9183 - val_loss: 0.6876 - val_accuracy: 0.7882\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.2670 - accuracy: 0.9238 - val_loss: 0.7785 - val_accuracy: 0.7560\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.2585 - accuracy: 0.9245 - val_loss: 0.6184 - val_accuracy: 0.8101\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.2419 - accuracy: 0.9310 - val_loss: 0.5494 - val_accuracy: 0.8306\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.2342 - accuracy: 0.9322 - val_loss: 0.6446 - val_accuracy: 0.7995\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.2206 - accuracy: 0.9367 - val_loss: 0.6844 - val_accuracy: 0.7965\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.2123 - accuracy: 0.9389 - val_loss: 0.6288 - val_accuracy: 0.8073\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.2074 - accuracy: 0.9396 - val_loss: 0.6681 - val_accuracy: 0.7950\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1948 - accuracy: 0.9444 - val_loss: 0.6732 - val_accuracy: 0.7978\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.1817 - accuracy: 0.9487 - val_loss: 0.5994 - val_accuracy: 0.8242\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1752 - accuracy: 0.9503 - val_loss: 0.8625 - val_accuracy: 0.7724\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1676 - accuracy: 0.9521 - val_loss: 0.6120 - val_accuracy: 0.8186\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1642 - accuracy: 0.9536 - val_loss: 0.7485 - val_accuracy: 0.7867\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 0.1612 - accuracy: 0.9535 - val_loss: 0.6007 - val_accuracy: 0.8207\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.1520 - accuracy: 0.9566 - val_loss: 0.6308 - val_accuracy: 0.8108\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1449 - accuracy: 0.9595 - val_loss: 0.6533 - val_accuracy: 0.8158\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.1410 - accuracy: 0.9598 - val_loss: 0.5447 - val_accuracy: 0.8456\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 9s 219us/step - loss: 0.1325 - accuracy: 0.9620 - val_loss: 0.6665 - val_accuracy: 0.8134\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.1295 - accuracy: 0.9641 - val_loss: 0.6574 - val_accuracy: 0.8123\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.1310 - accuracy: 0.9623 - val_loss: 0.6271 - val_accuracy: 0.8217\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1226 - accuracy: 0.9656 - val_loss: 0.6042 - val_accuracy: 0.8279\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1189 - accuracy: 0.9669 - val_loss: 0.9131 - val_accuracy: 0.7731\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.1173 - accuracy: 0.9668 - val_loss: 0.5728 - val_accuracy: 0.8341\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1090 - accuracy: 0.9697 - val_loss: 0.6074 - val_accuracy: 0.8311\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1074 - accuracy: 0.9701 - val_loss: 0.6233 - val_accuracy: 0.8227\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.1041 - accuracy: 0.9705 - val_loss: 0.5863 - val_accuracy: 0.8404\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.1014 - accuracy: 0.9714 - val_loss: 0.8470 - val_accuracy: 0.7788\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.1001 - accuracy: 0.9717 - val_loss: 0.7040 - val_accuracy: 0.8098\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0987 - accuracy: 0.9722 - val_loss: 0.8942 - val_accuracy: 0.7828\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0931 - accuracy: 0.9736 - val_loss: 0.7835 - val_accuracy: 0.7943\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0919 - accuracy: 0.9748 - val_loss: 0.6418 - val_accuracy: 0.8269\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0864 - accuracy: 0.9757 - val_loss: 0.7276 - val_accuracy: 0.8066\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.0925 - accuracy: 0.9731 - val_loss: 0.6468 - val_accuracy: 0.8332\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0826 - accuracy: 0.9766 - val_loss: 0.6038 - val_accuracy: 0.8375\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.0865 - accuracy: 0.9744 - val_loss: 0.5630 - val_accuracy: 0.8494\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0737 - accuracy: 0.9795 - val_loss: 0.7449 - val_accuracy: 0.8141\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0781 - accuracy: 0.9771 - val_loss: 0.6381 - val_accuracy: 0.8331\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0776 - accuracy: 0.9775 - val_loss: 0.7493 - val_accuracy: 0.8070\n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0746 - accuracy: 0.9785 - val_loss: 0.6026 - val_accuracy: 0.8376\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0744 - accuracy: 0.9784 - val_loss: 0.7114 - val_accuracy: 0.8203\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 9s 212us/step - loss: 0.0710 - accuracy: 0.9803 - val_loss: 0.6880 - val_accuracy: 0.8178\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0709 - accuracy: 0.9793 - val_loss: 0.8315 - val_accuracy: 0.7986\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0733 - accuracy: 0.9782 - val_loss: 0.5665 - val_accuracy: 0.8498\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0645 - accuracy: 0.9816 - val_loss: 0.6819 - val_accuracy: 0.8225\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0679 - accuracy: 0.9795 - val_loss: 0.8178 - val_accuracy: 0.8104\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0679 - accuracy: 0.9803 - val_loss: 0.5714 - val_accuracy: 0.8538\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0598 - accuracy: 0.9833 - val_loss: 0.7585 - val_accuracy: 0.8233\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.0617 - accuracy: 0.9822 - val_loss: 0.6117 - val_accuracy: 0.8438\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0688 - accuracy: 0.9795 - val_loss: 0.8847 - val_accuracy: 0.7858\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0648 - accuracy: 0.9809 - val_loss: 0.7933 - val_accuracy: 0.8094\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0679 - accuracy: 0.9794 - val_loss: 0.7962 - val_accuracy: 0.8106\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0585 - accuracy: 0.9822 - val_loss: 0.7509 - val_accuracy: 0.8182\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 9s 218us/step - loss: 0.0556 - accuracy: 0.9842 - val_loss: 0.8509 - val_accuracy: 0.8102\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0567 - accuracy: 0.9833 - val_loss: 0.7641 - val_accuracy: 0.8183\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0610 - accuracy: 0.9815 - val_loss: 0.7635 - val_accuracy: 0.8195\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0601 - accuracy: 0.9820 - val_loss: 0.7404 - val_accuracy: 0.8228\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0508 - accuracy: 0.9851 - val_loss: 0.6271 - val_accuracy: 0.8520\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 9s 216us/step - loss: 0.0519 - accuracy: 0.9851 - val_loss: 1.0091 - val_accuracy: 0.7713\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.0546 - accuracy: 0.9835 - val_loss: 0.7315 - val_accuracy: 0.8225\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0505 - accuracy: 0.9846 - val_loss: 0.5991 - val_accuracy: 0.8533\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 9s 225us/step - loss: 0.0484 - accuracy: 0.9867 - val_loss: 0.6747 - val_accuracy: 0.8367\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0562 - accuracy: 0.9828 - val_loss: 0.8096 - val_accuracy: 0.8044\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 9s 215us/step - loss: 0.0471 - accuracy: 0.9867 - val_loss: 0.5981 - val_accuracy: 0.8568\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 9s 220us/step - loss: 0.0472 - accuracy: 0.9859 - val_loss: 0.6778 - val_accuracy: 0.8344\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0477 - accuracy: 0.9855 - val_loss: 0.6950 - val_accuracy: 0.8321\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 10s 226us/step - loss: 0.0604 - accuracy: 0.9815 - val_loss: 1.3955 - val_accuracy: 0.7383\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0504 - accuracy: 0.9845 - val_loss: 0.7784 - val_accuracy: 0.8268\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0519 - accuracy: 0.9842 - val_loss: 0.7211 - val_accuracy: 0.8266\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 10s 226us/step - loss: 0.0470 - accuracy: 0.9861 - val_loss: 0.8596 - val_accuracy: 0.8030\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0490 - accuracy: 0.9858 - val_loss: 0.6548 - val_accuracy: 0.8450\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 9s 217us/step - loss: 0.0510 - accuracy: 0.9840 - val_loss: 0.6632 - val_accuracy: 0.8455\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 9s 225us/step - loss: 0.0410 - accuracy: 0.9881 - val_loss: 0.7258 - val_accuracy: 0.8260\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 0.0456 - accuracy: 0.9866 - val_loss: 1.1241 - val_accuracy: 0.7663\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 9s 214us/step - loss: 0.0483 - accuracy: 0.9856 - val_loss: 0.8132 - val_accuracy: 0.8273\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 9s 221us/step - loss: 0.0421 - accuracy: 0.9874 - val_loss: 0.7501 - val_accuracy: 0.8382\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 9s 213us/step - loss: 0.0422 - accuracy: 0.9878 - val_loss: 0.5412 - val_accuracy: 0.8749\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 9s 221us/step - loss: 0.0463 - accuracy: 0.9861 - val_loss: 0.6164 - val_accuracy: 0.8526\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.0457 - accuracy: 0.9860 - val_loss: 0.5840 - val_accuracy: 0.8680\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history_2 = keras_model_2.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 321,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60000/60000 [==============================] - 7s 114us/step\n",
      "Val_acc using BatchNorm sgd :  0.8679500222206116\n"
     ]
    }
   ],
   "source": [
    "results_2 = keras_model_2.evaluate(X_val, y_val)\n",
    "print('Val_acc using BatchNorm sgd : ', results_2[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 322,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Method</th>\n",
       "      <th>accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>NN</td>\n",
       "      <td>0.826400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>NN_Batch_adam</td>\n",
       "      <td>0.831467</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>NN_Batch_sgd</td>\n",
       "      <td>0.867950</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          Method  accuracy\n",
       "1             NN  0.826400\n",
       "2  NN_Batch_adam  0.831467\n",
       "3   NN_Batch_sgd  0.867950"
      ]
     },
     "execution_count": 322,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Store the accuracy results for each model in a dataframe for final comparison\n",
    "tempResultsDf = pd.DataFrame({'Method':['NN_Batch_sgd'], 'accuracy': [results_2[1]]},index={'3'})\n",
    "results_on_val= pd.concat([results_on_val, tempResultsDf])\n",
    "results_on_val = results_on_val[['Method', 'accuracy']]\n",
    "results_on_val"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* The batch normalization using SGD optimizer is giving better results than using adam optimizer."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Implementing batch normalization and Dropout for training the neural network "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 352,
   "metadata": {},
   "outputs": [],
   "source": [
    "##Initialize the Artificial Neural Network Classifier\n",
    "keras_model_3 = Sequential()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Adding BatchNormalization Layer and Dropout Layer (0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 353,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Input Layer\n",
    "#Adding Input layer and activation functions ReLU\n",
    "keras_model_3.add(Dense(512, kernel_initializer='he_normal',input_shape = (1024, )))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_3.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_3.add(Activation('relu'))\n",
    "#Adding Dropout Layer\n",
    "keras_model_3.add(Dropout(0.5))\n",
    "\n",
    "#Hidden Layer 1\n",
    "#Adding first Hidden layer\n",
    "keras_model_3.add(Dense(256,kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_3.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_3.add(Activation('relu'))\n",
    "#Adding Dropout Layer\n",
    "keras_model_3.add(Dropout(0.5))\n",
    "\n",
    "#Hidden Layer 2\n",
    "#Adding second Hidden layer\n",
    "keras_model_3.add(Dense(128, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_3.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_3.add(Activation('relu'))\n",
    "#Adding Dropout Layer\n",
    "keras_model_3.add(Dropout(0.5))\n",
    "\n",
    "#Hidden Layer 3\n",
    "#Adding third Hidden layer\n",
    "keras_model_3.add(Dense(64, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_3.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_3.add(Activation('relu'))\n",
    "#Adding Dropout Layer\n",
    "keras_model_3.add(Dropout(0.5))\n",
    "\n",
    "#Hidden Layer 4\n",
    "#Adding fourth Hidden layer\n",
    "keras_model_3.add(Dense(32, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_3.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_3.add(Activation('relu'))\n",
    "#Adding Dropout Layer\n",
    "keras_model_3.add(Dropout(0.5))\n",
    "\n",
    "# Output Layer\n",
    "#Adding output layer which is of 10 nodes (digidts)\n",
    "keras_model_3.add(Dense(10))\n",
    "#Adding Activation function\n",
    "# Here, we are using softmax function because we have multiclass classsification\n",
    "keras_model_3.add(Activation('softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 339,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_6\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_31 (Dense)             (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "batch_normalization_16 (Batc (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "activation_31 (Activation)   (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dropout_6 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_32 (Dense)             (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "batch_normalization_17 (Batc (None, 256)               1024      \n",
      "_________________________________________________________________\n",
      "activation_32 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dropout_7 (Dropout)          (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_33 (Dense)             (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "batch_normalization_18 (Batc (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_33 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_8 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_34 (Dense)             (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "batch_normalization_19 (Batc (None, 64)                256       \n",
      "_________________________________________________________________\n",
      "activation_34 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dropout_9 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_35 (Dense)             (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "batch_normalization_20 (Batc (None, 32)                128       \n",
      "_________________________________________________________________\n",
      "activation_35 (Activation)   (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dropout_10 (Dropout)         (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_36 (Dense)             (None, 10)                330       \n",
      "_________________________________________________________________\n",
      "activation_36 (Activation)   (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 703,658\n",
      "Trainable params: 701,674\n",
      "Non-trainable params: 1,984\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras_model_3.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Using ADAM optimizer.\n",
    "* learning rate= 0.0001"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 340,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Setting the learning rate in adam\n",
    "adam = optimizers.adam(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model_3.compile(optimizer = adam, loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 341,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 60000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 11s 264us/step - loss: 2.7309 - accuracy: 0.0985 - val_loss: 2.3234 - val_accuracy: 0.1142\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 2.6256 - accuracy: 0.1015 - val_loss: 2.3176 - val_accuracy: 0.1157\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 2.5558 - accuracy: 0.1025 - val_loss: 2.3059 - val_accuracy: 0.1110\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 2.4861 - accuracy: 0.1036 - val_loss: 2.3017 - val_accuracy: 0.1034\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 10s 240us/step - loss: 2.4489 - accuracy: 0.1036 - val_loss: 2.2963 - val_accuracy: 0.1119\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 10s 231us/step - loss: 2.4209 - accuracy: 0.1050 - val_loss: 2.2972 - val_accuracy: 0.1169\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 10s 239us/step - loss: 2.3986 - accuracy: 0.1036 - val_loss: 2.2949 - val_accuracy: 0.1226\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.3751 - accuracy: 0.1052 - val_loss: 2.2938 - val_accuracy: 0.1248\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 10s 244us/step - loss: 2.3660 - accuracy: 0.1039 - val_loss: 2.2938 - val_accuracy: 0.1408\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.3538 - accuracy: 0.1019 - val_loss: 2.2936 - val_accuracy: 0.1542\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.3416 - accuracy: 0.1073 - val_loss: 2.2910 - val_accuracy: 0.1754\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.3370 - accuracy: 0.1053 - val_loss: 2.2901 - val_accuracy: 0.1877\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.3295 - accuracy: 0.1083 - val_loss: 2.2875 - val_accuracy: 0.1962\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.3234 - accuracy: 0.1069 - val_loss: 2.2834 - val_accuracy: 0.2068\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.3154 - accuracy: 0.1111 - val_loss: 2.2792 - val_accuracy: 0.1893\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.3095 - accuracy: 0.1149 - val_loss: 2.2716 - val_accuracy: 0.2009\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.3035 - accuracy: 0.1205 - val_loss: 2.2634 - val_accuracy: 0.1944\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.2917 - accuracy: 0.1238 - val_loss: 2.2411 - val_accuracy: 0.2193\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.2784 - accuracy: 0.1349 - val_loss: 2.2273 - val_accuracy: 0.1978\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 2.2618 - accuracy: 0.1418 - val_loss: 2.1955 - val_accuracy: 0.2044\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 2.2378 - accuracy: 0.1537 - val_loss: 2.1608 - val_accuracy: 0.2076\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.2139 - accuracy: 0.1577 - val_loss: 2.1228 - val_accuracy: 0.2104\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.1806 - accuracy: 0.1708 - val_loss: 2.0804 - val_accuracy: 0.2095\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.1493 - accuracy: 0.1770 - val_loss: 2.0761 - val_accuracy: 0.2092\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.1203 - accuracy: 0.1868 - val_loss: 2.0017 - val_accuracy: 0.2550\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.0889 - accuracy: 0.1945 - val_loss: 1.9579 - val_accuracy: 0.2978\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 2.0625 - accuracy: 0.2059 - val_loss: 1.9345 - val_accuracy: 0.3075\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 2.0245 - accuracy: 0.2178 - val_loss: 1.8919 - val_accuracy: 0.3113\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 1.9894 - accuracy: 0.2297 - val_loss: 1.8293 - val_accuracy: 0.3494\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 10s 240us/step - loss: 1.9594 - accuracy: 0.2458 - val_loss: 1.8302 - val_accuracy: 0.3180\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 10s 242us/step - loss: 1.9267 - accuracy: 0.2580 - val_loss: 1.7587 - val_accuracy: 0.3806\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 10s 244us/step - loss: 1.8933 - accuracy: 0.2709 - val_loss: 1.7414 - val_accuracy: 0.3795\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 1.8660 - accuracy: 0.2824 - val_loss: 1.7095 - val_accuracy: 0.3813\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 1.8357 - accuracy: 0.2921 - val_loss: 1.7024 - val_accuracy: 0.3857\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 10s 234us/step - loss: 1.8093 - accuracy: 0.3059 - val_loss: 1.6372 - val_accuracy: 0.4319\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 1.7771 - accuracy: 0.3196 - val_loss: 1.5753 - val_accuracy: 0.4683\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 10s 233us/step - loss: 1.7482 - accuracy: 0.3370 - val_loss: 1.5557 - val_accuracy: 0.4653\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 10s 242us/step - loss: 1.7236 - accuracy: 0.3475 - val_loss: 1.4725 - val_accuracy: 0.5150\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.6941 - accuracy: 0.3558 - val_loss: 1.4344 - val_accuracy: 0.5244\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 10s 235us/step - loss: 1.6628 - accuracy: 0.3753 - val_loss: 1.4363 - val_accuracy: 0.5128\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 10s 239us/step - loss: 1.6397 - accuracy: 0.3798 - val_loss: 1.4255 - val_accuracy: 0.5275\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 10s 243us/step - loss: 1.6085 - accuracy: 0.3920 - val_loss: 1.3536 - val_accuracy: 0.5473\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 10s 242us/step - loss: 1.5855 - accuracy: 0.4062 - val_loss: 1.3434 - val_accuracy: 0.5645\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 10s 241us/step - loss: 1.5676 - accuracy: 0.4094 - val_loss: 1.3461 - val_accuracy: 0.5420\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 11s 251us/step - loss: 1.5443 - accuracy: 0.4167 - val_loss: 1.3148 - val_accuracy: 0.5174\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 10s 246us/step - loss: 1.5282 - accuracy: 0.4239 - val_loss: 1.2499 - val_accuracy: 0.5779\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 10s 247us/step - loss: 1.5084 - accuracy: 0.4310 - val_loss: 1.2551 - val_accuracy: 0.5896\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 11s 256us/step - loss: 1.4957 - accuracy: 0.4320 - val_loss: 1.2622 - val_accuracy: 0.5842\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 11s 255us/step - loss: 1.4850 - accuracy: 0.4380 - val_loss: 1.2330 - val_accuracy: 0.5721\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 11s 255us/step - loss: 1.4656 - accuracy: 0.4499 - val_loss: 1.2973 - val_accuracy: 0.5633\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 11s 255us/step - loss: 1.4604 - accuracy: 0.4494 - val_loss: 1.2099 - val_accuracy: 0.6121\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 11s 259us/step - loss: 1.4426 - accuracy: 0.4576 - val_loss: 1.1872 - val_accuracy: 0.6216\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 11s 258us/step - loss: 1.4320 - accuracy: 0.4638 - val_loss: 1.2299 - val_accuracy: 0.5559\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 10s 250us/step - loss: 1.4167 - accuracy: 0.4693 - val_loss: 1.1535 - val_accuracy: 0.6348\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 11s 250us/step - loss: 1.4056 - accuracy: 0.4741 - val_loss: 1.1342 - val_accuracy: 0.6323\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 1.3949 - accuracy: 0.4786 - val_loss: 1.2140 - val_accuracy: 0.6209\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 10s 236us/step - loss: 1.3859 - accuracy: 0.4847 - val_loss: 1.1345 - val_accuracy: 0.6511\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 1.3802 - accuracy: 0.4839 - val_loss: 1.1354 - val_accuracy: 0.6396\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 1.3710 - accuracy: 0.4891 - val_loss: 1.0938 - val_accuracy: 0.6623\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 1.3582 - accuracy: 0.4962 - val_loss: 1.1002 - val_accuracy: 0.6469\n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 10s 240us/step - loss: 1.3475 - accuracy: 0.5036 - val_loss: 1.1241 - val_accuracy: 0.6478\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 10s 238us/step - loss: 1.3386 - accuracy: 0.5056 - val_loss: 1.0762 - val_accuracy: 0.6650\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 1.3295 - accuracy: 0.5106 - val_loss: 1.0630 - val_accuracy: 0.6853\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 1.3221 - accuracy: 0.5145 - val_loss: 1.0686 - val_accuracy: 0.6728\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 10s 244us/step - loss: 1.3113 - accuracy: 0.5204 - val_loss: 1.0825 - val_accuracy: 0.6614\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 10s 241us/step - loss: 1.3066 - accuracy: 0.5244 - val_loss: 1.0108 - val_accuracy: 0.7016\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 10s 240us/step - loss: 1.2987 - accuracy: 0.5304 - val_loss: 1.0167 - val_accuracy: 0.6908\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 10s 242us/step - loss: 1.2896 - accuracy: 0.5332 - val_loss: 1.0609 - val_accuracy: 0.6594\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 10s 244us/step - loss: 1.2910 - accuracy: 0.5328 - val_loss: 1.0157 - val_accuracy: 0.6990\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 10s 239us/step - loss: 1.2812 - accuracy: 0.5369 - val_loss: 1.0315 - val_accuracy: 0.6869\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 10s 235us/step - loss: 1.2701 - accuracy: 0.5473 - val_loss: 1.0893 - val_accuracy: 0.6447\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 10s 242us/step - loss: 1.2595 - accuracy: 0.5465 - val_loss: 1.0147 - val_accuracy: 0.6582\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 1.2541 - accuracy: 0.5578 - val_loss: 0.9983 - val_accuracy: 0.6981\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 1.2496 - accuracy: 0.5558 - val_loss: 0.9794 - val_accuracy: 0.6962\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 10s 229us/step - loss: 1.2360 - accuracy: 0.5610 - val_loss: 0.9760 - val_accuracy: 0.7022\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 11s 255us/step - loss: 1.2311 - accuracy: 0.5658 - val_loss: 1.0183 - val_accuracy: 0.6880\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.2224 - accuracy: 0.5665 - val_loss: 0.9428 - val_accuracy: 0.7116\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.2169 - accuracy: 0.5718 - val_loss: 0.9264 - val_accuracy: 0.7311\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 10s 243us/step - loss: 1.2085 - accuracy: 0.5756 - val_loss: 0.9439 - val_accuracy: 0.6960\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 10s 248us/step - loss: 1.2008 - accuracy: 0.5811 - val_loss: 0.9186 - val_accuracy: 0.7149\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 10s 246us/step - loss: 1.1987 - accuracy: 0.5847 - val_loss: 0.9537 - val_accuracy: 0.6997\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.1906 - accuracy: 0.5840 - val_loss: 0.9212 - val_accuracy: 0.7236\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 12s 292us/step - loss: 1.1844 - accuracy: 0.5887 - val_loss: 0.9020 - val_accuracy: 0.7289\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 10s 235us/step - loss: 1.1806 - accuracy: 0.5889 - val_loss: 0.9427 - val_accuracy: 0.6910\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 10s 243us/step - loss: 1.1719 - accuracy: 0.5950 - val_loss: 0.8913 - val_accuracy: 0.7352\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 10s 237us/step - loss: 1.1642 - accuracy: 0.5999 - val_loss: 0.9248 - val_accuracy: 0.6988\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 10s 247us/step - loss: 1.1619 - accuracy: 0.6011 - val_loss: 0.9185 - val_accuracy: 0.7229\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 11s 256us/step - loss: 1.1526 - accuracy: 0.6059 - val_loss: 0.8842 - val_accuracy: 0.7285\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 10s 243us/step - loss: 1.1465 - accuracy: 0.6081 - val_loss: 0.8897 - val_accuracy: 0.7229\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 11s 252us/step - loss: 1.1350 - accuracy: 0.6104 - val_loss: 0.8616 - val_accuracy: 0.7376\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 10s 243us/step - loss: 1.1320 - accuracy: 0.6136 - val_loss: 0.8573 - val_accuracy: 0.7329\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 11s 253us/step - loss: 1.1334 - accuracy: 0.6142 - val_loss: 0.8900 - val_accuracy: 0.7251\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 10s 227us/step - loss: 1.1279 - accuracy: 0.6187 - val_loss: 0.8452 - val_accuracy: 0.7412\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 11s 251us/step - loss: 1.1138 - accuracy: 0.6227 - val_loss: 0.8699 - val_accuracy: 0.7258\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.1111 - accuracy: 0.6253 - val_loss: 0.8483 - val_accuracy: 0.7342\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 10s 240us/step - loss: 1.1130 - accuracy: 0.6256 - val_loss: 0.9260 - val_accuracy: 0.6981\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 10s 247us/step - loss: 1.1069 - accuracy: 0.6278 - val_loss: 0.8520 - val_accuracy: 0.7316\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 10s 245us/step - loss: 1.0987 - accuracy: 0.6302 - val_loss: 0.8299 - val_accuracy: 0.7434\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 11s 253us/step - loss: 1.0970 - accuracy: 0.6314 - val_loss: 0.8114 - val_accuracy: 0.7505\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 10s 236us/step - loss: 1.0880 - accuracy: 0.6364 - val_loss: 0.8478 - val_accuracy: 0.7424\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history_3 = keras_model_3.fit(X_train, y_train, validation_data=(X_val,y_val),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 342,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60000/60000 [==============================] - 8s 127us/step\n",
      "Val_acc using BatchNorm and Dropout adam :  0.7423833608627319\n"
     ]
    }
   ],
   "source": [
    "results_3 = keras_model_3.evaluate(X_val, y_val)\n",
    "print('Val_acc using BatchNorm and Dropout adam : ', results_3[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 343,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Method</th>\n",
       "      <th>accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>NN</td>\n",
       "      <td>0.826400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>NN_Batch_adam</td>\n",
       "      <td>0.831467</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>NN_Batch_sgd</td>\n",
       "      <td>0.867950</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>NN_Batch_Drop_adam</td>\n",
       "      <td>0.742383</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               Method  accuracy\n",
       "1                  NN  0.826400\n",
       "2       NN_Batch_adam  0.831467\n",
       "3        NN_Batch_sgd  0.867950\n",
       "4  NN_Batch_Drop_adam  0.742383"
      ]
     },
     "execution_count": 343,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Store the accuracy results for each model in a dataframe for final comparison\n",
    "tempResultsDf = pd.DataFrame({'Method':['NN_Batch_Drop_adam'], 'accuracy': [results_3[1]]},index={'4'})\n",
    "results_on_val= pd.concat([results_on_val, tempResultsDf])\n",
    "results_on_val = results_on_val[['Method', 'accuracy']]\n",
    "results_on_val"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing NN on test data "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "##Initialize the Artificial Neural Network Classifier\n",
    "keras_model_final = Sequential()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Input Layer\n",
    "#Adding Input layer and activation functions ReLU\n",
    "keras_model_final.add(Dense(512, kernel_initializer='he_normal',input_shape = (1024, )))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_final.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_final.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 1\n",
    "#Adding first Hidden layer\n",
    "keras_model_final.add(Dense(256, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_final.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_final.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 2\n",
    "#Adding second Hidden layer\n",
    "keras_model_final.add(Dense(128, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_final.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_final.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 3\n",
    "#Adding third Hidden layer\n",
    "keras_model_final.add(Dense(64, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_final.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_final.add(Activation('relu'))\n",
    "\n",
    "#Hidden Layer 4\n",
    "#Adding fourth Hidden layer\n",
    "keras_model_final.add(Dense(32, kernel_initializer='he_normal'))\n",
    "#Adding BatchNormalization Layer\n",
    "keras_model_final.add(BatchNormalization())\n",
    "#Adding Activation function\n",
    "keras_model_final.add(Activation('relu'))\n",
    "\n",
    "# Output Layer\n",
    "#Adding output layer which is of 10 nodes (digidts)\n",
    "keras_model_final.add(Dense(10))\n",
    "#Adding Activation function\n",
    "# Here, we are using softmax function because we have multiclass classsification\n",
    "keras_model_final.add(Activation('softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_3\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_13 (Dense)             (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "batch_normalization_11 (Batc (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "activation_13 (Activation)   (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_14 (Dense)             (None, 256)               131328    \n",
      "_________________________________________________________________\n",
      "batch_normalization_12 (Batc (None, 256)               1024      \n",
      "_________________________________________________________________\n",
      "activation_14 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_15 (Dense)             (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "batch_normalization_13 (Batc (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "activation_15 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_16 (Dense)             (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "batch_normalization_14 (Batc (None, 64)                256       \n",
      "_________________________________________________________________\n",
      "activation_16 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_17 (Dense)             (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "batch_normalization_15 (Batc (None, 32)                128       \n",
      "_________________________________________________________________\n",
      "activation_17 (Activation)   (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_18 (Dense)             (None, 10)                330       \n",
      "_________________________________________________________________\n",
      "activation_18 (Activation)   (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 703,658\n",
      "Trainable params: 701,674\n",
      "Non-trainable params: 1,984\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras_model_final.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Setting the learning rate in sgd\n",
    "sgd = optimizers.adam(lr = 0.0001)\n",
    "# compiling the ANN classifier\n",
    "keras_model_final.compile(optimizer = sgd, loss = 'categorical_crossentropy', metrics = ['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 42000 samples, validate on 18000 samples\n",
      "Epoch 1/100\n",
      "42000/42000 [==============================] - 8s 182us/step - loss: 2.1438 - accuracy: 0.2544 - val_loss: 2.0569 - val_accuracy: 0.2821\n",
      "Epoch 2/100\n",
      "42000/42000 [==============================] - 6s 150us/step - loss: 1.5995 - accuracy: 0.5087 - val_loss: 1.6682 - val_accuracy: 0.4636\n",
      "Epoch 3/100\n",
      "42000/42000 [==============================] - 7s 177us/step - loss: 1.3022 - accuracy: 0.6312 - val_loss: 1.3960 - val_accuracy: 0.5786\n",
      "Epoch 4/100\n",
      "42000/42000 [==============================] - 8s 201us/step - loss: 1.1083 - accuracy: 0.6922 - val_loss: 1.2876 - val_accuracy: 0.6150\n",
      "Epoch 5/100\n",
      "42000/42000 [==============================] - 8s 188us/step - loss: 0.9710 - accuracy: 0.7306 - val_loss: 1.1591 - val_accuracy: 0.6525\n",
      "Epoch 6/100\n",
      "42000/42000 [==============================] - 9s 208us/step - loss: 0.8685 - accuracy: 0.7535 - val_loss: 1.1056 - val_accuracy: 0.6734\n",
      "Epoch 7/100\n",
      "42000/42000 [==============================] - 8s 185us/step - loss: 0.7831 - accuracy: 0.7772 - val_loss: 1.0700 - val_accuracy: 0.6727\n",
      "Epoch 8/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.7168 - accuracy: 0.7956 - val_loss: 1.0123 - val_accuracy: 0.6864\n",
      "Epoch 9/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.6585 - accuracy: 0.8086 - val_loss: 0.9798 - val_accuracy: 0.6983\n",
      "Epoch 10/100\n",
      "42000/42000 [==============================] - 8s 196us/step - loss: 0.6064 - accuracy: 0.8247 - val_loss: 1.0498 - val_accuracy: 0.6719\n",
      "Epoch 11/100\n",
      "42000/42000 [==============================] - 10s 232us/step - loss: 0.5626 - accuracy: 0.8370 - val_loss: 0.9504 - val_accuracy: 0.7032\n",
      "Epoch 12/100\n",
      "42000/42000 [==============================] - 9s 223us/step - loss: 0.5234 - accuracy: 0.8480 - val_loss: 0.8432 - val_accuracy: 0.7419\n",
      "Epoch 13/100\n",
      "42000/42000 [==============================] - 8s 191us/step - loss: 0.4915 - accuracy: 0.8575 - val_loss: 0.9291 - val_accuracy: 0.7112\n",
      "Epoch 14/100\n",
      "42000/42000 [==============================] - 8s 190us/step - loss: 0.4652 - accuracy: 0.8632 - val_loss: 0.9583 - val_accuracy: 0.7039\n",
      "Epoch 15/100\n",
      "42000/42000 [==============================] - 8s 185us/step - loss: 0.4346 - accuracy: 0.8739 - val_loss: 0.9642 - val_accuracy: 0.7054\n",
      "Epoch 16/100\n",
      "42000/42000 [==============================] - 10s 226us/step - loss: 0.4132 - accuracy: 0.8797 - val_loss: 0.9794 - val_accuracy: 0.6993\n",
      "Epoch 17/100\n",
      "42000/42000 [==============================] - 8s 187us/step - loss: 0.3903 - accuracy: 0.8857 - val_loss: 0.9589 - val_accuracy: 0.7097\n",
      "Epoch 18/100\n",
      "42000/42000 [==============================] - 8s 187us/step - loss: 0.3704 - accuracy: 0.8919 - val_loss: 0.8512 - val_accuracy: 0.7422\n",
      "Epoch 19/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.3491 - accuracy: 0.8987 - val_loss: 1.1295 - val_accuracy: 0.6747\n",
      "Epoch 20/100\n",
      "42000/42000 [==============================] - 8s 185us/step - loss: 0.3316 - accuracy: 0.9050 - val_loss: 0.9389 - val_accuracy: 0.7133\n",
      "Epoch 21/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.3139 - accuracy: 0.9094 - val_loss: 0.8345 - val_accuracy: 0.7466\n",
      "Epoch 22/100\n",
      "42000/42000 [==============================] - 8s 187us/step - loss: 0.2981 - accuracy: 0.9156 - val_loss: 0.8607 - val_accuracy: 0.7408\n",
      "Epoch 23/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.2849 - accuracy: 0.9183 - val_loss: 0.9354 - val_accuracy: 0.7225\n",
      "Epoch 24/100\n",
      "42000/42000 [==============================] - 8s 186us/step - loss: 0.2759 - accuracy: 0.9197 - val_loss: 0.9014 - val_accuracy: 0.7341\n",
      "Epoch 25/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.2564 - accuracy: 0.9270 - val_loss: 0.9469 - val_accuracy: 0.7269\n",
      "Epoch 26/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.2482 - accuracy: 0.9290 - val_loss: 0.9456 - val_accuracy: 0.7230\n",
      "Epoch 27/100\n",
      "42000/42000 [==============================] - 9s 211us/step - loss: 0.2357 - accuracy: 0.9321 - val_loss: 0.9550 - val_accuracy: 0.7262\n",
      "Epoch 28/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.2246 - accuracy: 0.9359 - val_loss: 0.9726 - val_accuracy: 0.7253\n",
      "Epoch 29/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.2171 - accuracy: 0.9386 - val_loss: 0.9302 - val_accuracy: 0.7384\n",
      "Epoch 30/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.2058 - accuracy: 0.9423 - val_loss: 1.1729 - val_accuracy: 0.6899\n",
      "Epoch 31/100\n",
      "42000/42000 [==============================] - 8s 194us/step - loss: 0.2037 - accuracy: 0.9414 - val_loss: 0.9346 - val_accuracy: 0.7389\n",
      "Epoch 32/100\n",
      "42000/42000 [==============================] - 8s 186us/step - loss: 0.1905 - accuracy: 0.9452 - val_loss: 0.8721 - val_accuracy: 0.7564\n",
      "Epoch 33/100\n",
      "42000/42000 [==============================] - 8s 185us/step - loss: 0.1823 - accuracy: 0.9482 - val_loss: 1.0638 - val_accuracy: 0.7147\n",
      "Epoch 34/100\n",
      "42000/42000 [==============================] - 9s 204us/step - loss: 0.1748 - accuracy: 0.9505 - val_loss: 0.9655 - val_accuracy: 0.7353\n",
      "Epoch 35/100\n",
      "42000/42000 [==============================] - 7s 177us/step - loss: 0.1650 - accuracy: 0.9526 - val_loss: 0.9948 - val_accuracy: 0.7346\n",
      "Epoch 36/100\n",
      "42000/42000 [==============================] - 7s 169us/step - loss: 0.1634 - accuracy: 0.9542 - val_loss: 1.0044 - val_accuracy: 0.7334\n",
      "Epoch 37/100\n",
      "42000/42000 [==============================] - 6s 142us/step - loss: 0.1591 - accuracy: 0.9543 - val_loss: 0.9090 - val_accuracy: 0.7498\n",
      "Epoch 38/100\n",
      "42000/42000 [==============================] - 6s 144us/step - loss: 0.1520 - accuracy: 0.9576 - val_loss: 1.0260 - val_accuracy: 0.7363\n",
      "Epoch 39/100\n",
      "42000/42000 [==============================] - 6s 137us/step - loss: 0.1458 - accuracy: 0.9590 - val_loss: 0.9881 - val_accuracy: 0.7406\n",
      "Epoch 40/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.1439 - accuracy: 0.9593 - val_loss: 1.1461 - val_accuracy: 0.7086\n",
      "Epoch 41/100\n",
      "42000/42000 [==============================] - 7s 162us/step - loss: 0.1361 - accuracy: 0.9613 - val_loss: 0.9838 - val_accuracy: 0.7448\n",
      "Epoch 42/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.1288 - accuracy: 0.9644 - val_loss: 1.0964 - val_accuracy: 0.7185\n",
      "Epoch 43/100\n",
      "42000/42000 [==============================] - 6s 144us/step - loss: 0.1294 - accuracy: 0.9625 - val_loss: 1.0307 - val_accuracy: 0.7381\n",
      "Epoch 44/100\n",
      "42000/42000 [==============================] - 6s 140us/step - loss: 0.1226 - accuracy: 0.9650 - val_loss: 1.0973 - val_accuracy: 0.7221\n",
      "Epoch 45/100\n",
      "42000/42000 [==============================] - 6s 153us/step - loss: 0.1194 - accuracy: 0.9666 - val_loss: 1.0355 - val_accuracy: 0.7384\n",
      "Epoch 46/100\n",
      "42000/42000 [==============================] - 8s 200us/step - loss: 0.1191 - accuracy: 0.9657 - val_loss: 0.9397 - val_accuracy: 0.7610\n",
      "Epoch 47/100\n",
      "42000/42000 [==============================] - 7s 165us/step - loss: 0.1082 - accuracy: 0.9700 - val_loss: 1.1529 - val_accuracy: 0.7163\n",
      "Epoch 48/100\n",
      "42000/42000 [==============================] - 6s 140us/step - loss: 0.1087 - accuracy: 0.9689 - val_loss: 1.0105 - val_accuracy: 0.7475\n",
      "Epoch 49/100\n",
      "42000/42000 [==============================] - 6s 144us/step - loss: 0.1005 - accuracy: 0.9715 - val_loss: 1.4625 - val_accuracy: 0.6816\n",
      "Epoch 50/100\n",
      "42000/42000 [==============================] - 8s 191us/step - loss: 0.1077 - accuracy: 0.9688 - val_loss: 1.1563 - val_accuracy: 0.7252\n",
      "Epoch 51/100\n",
      "42000/42000 [==============================] - 8s 190us/step - loss: 0.0990 - accuracy: 0.9724 - val_loss: 1.2120 - val_accuracy: 0.7193\n",
      "Epoch 52/100\n",
      "42000/42000 [==============================] - 8s 190us/step - loss: 0.0977 - accuracy: 0.9717 - val_loss: 1.0556 - val_accuracy: 0.7424\n",
      "Epoch 53/100\n",
      "42000/42000 [==============================] - 8s 198us/step - loss: 0.0991 - accuracy: 0.9714 - val_loss: 0.9489 - val_accuracy: 0.7642\n",
      "Epoch 54/100\n",
      "42000/42000 [==============================] - 7s 156us/step - loss: 0.0901 - accuracy: 0.9746 - val_loss: 1.1530 - val_accuracy: 0.7309\n",
      "Epoch 55/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.0846 - accuracy: 0.9769 - val_loss: 1.0824 - val_accuracy: 0.7423\n",
      "Epoch 56/100\n",
      "42000/42000 [==============================] - 6s 140us/step - loss: 0.0917 - accuracy: 0.9727 - val_loss: 1.3240 - val_accuracy: 0.7076\n",
      "Epoch 57/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.0826 - accuracy: 0.9765 - val_loss: 1.1699 - val_accuracy: 0.7266\n",
      "Epoch 58/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.0918 - accuracy: 0.9729 - val_loss: 1.1174 - val_accuracy: 0.7341\n",
      "Epoch 59/100\n",
      "42000/42000 [==============================] - 6s 140us/step - loss: 0.0809 - accuracy: 0.9770 - val_loss: 1.1578 - val_accuracy: 0.7364\n",
      "Epoch 60/100\n",
      "42000/42000 [==============================] - 6s 150us/step - loss: 0.0779 - accuracy: 0.9777 - val_loss: 1.0567 - val_accuracy: 0.7583\n",
      "Epoch 61/100\n",
      "42000/42000 [==============================] - 6s 145us/step - loss: 0.0769 - accuracy: 0.9780 - val_loss: 1.1190 - val_accuracy: 0.7433\n",
      "Epoch 62/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.0759 - accuracy: 0.9779 - val_loss: 1.2433 - val_accuracy: 0.7279\n",
      "Epoch 63/100\n",
      "42000/42000 [==============================] - 6s 139us/step - loss: 0.0804 - accuracy: 0.9760 - val_loss: 1.4445 - val_accuracy: 0.7170\n",
      "Epoch 64/100\n",
      "42000/42000 [==============================] - 6s 139us/step - loss: 0.0784 - accuracy: 0.9768 - val_loss: 1.2017 - val_accuracy: 0.7337\n",
      "Epoch 65/100\n",
      "42000/42000 [==============================] - 6s 137us/step - loss: 0.0696 - accuracy: 0.9804 - val_loss: 1.2219 - val_accuracy: 0.7356\n",
      "Epoch 66/100\n",
      "42000/42000 [==============================] - 7s 168us/step - loss: 0.0703 - accuracy: 0.9801 - val_loss: 1.1661 - val_accuracy: 0.7462\n",
      "Epoch 67/100\n",
      "42000/42000 [==============================] - 8s 182us/step - loss: 0.0676 - accuracy: 0.9804 - val_loss: 1.3038 - val_accuracy: 0.7158\n",
      "Epoch 68/100\n",
      "42000/42000 [==============================] - 8s 182us/step - loss: 0.0691 - accuracy: 0.9798 - val_loss: 1.2912 - val_accuracy: 0.7199\n",
      "Epoch 69/100\n",
      "42000/42000 [==============================] - 8s 188us/step - loss: 0.0574 - accuracy: 0.9839 - val_loss: 1.2537 - val_accuracy: 0.7314\n",
      "Epoch 70/100\n",
      "42000/42000 [==============================] - 8s 185us/step - loss: 0.0686 - accuracy: 0.9799 - val_loss: 1.0932 - val_accuracy: 0.7477\n",
      "Epoch 71/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.0667 - accuracy: 0.9807 - val_loss: 1.3489 - val_accuracy: 0.7211\n",
      "Epoch 72/100\n",
      "42000/42000 [==============================] - 8s 183us/step - loss: 0.0642 - accuracy: 0.9816 - val_loss: 1.1751 - val_accuracy: 0.7368\n",
      "Epoch 73/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.0608 - accuracy: 0.9823 - val_loss: 1.1599 - val_accuracy: 0.7468\n",
      "Epoch 74/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.0632 - accuracy: 0.9816 - val_loss: 1.1890 - val_accuracy: 0.7451\n",
      "Epoch 75/100\n",
      "42000/42000 [==============================] - 8s 184us/step - loss: 0.0533 - accuracy: 0.9845 - val_loss: 1.2356 - val_accuracy: 0.7502\n",
      "Epoch 76/100\n",
      "42000/42000 [==============================] - 8s 196us/step - loss: 0.0577 - accuracy: 0.9833 - val_loss: 1.2271 - val_accuracy: 0.7474\n",
      "Epoch 77/100\n",
      "42000/42000 [==============================] - 8s 190us/step - loss: 0.0582 - accuracy: 0.9832 - val_loss: 1.3087 - val_accuracy: 0.7278\n",
      "Epoch 78/100\n",
      "42000/42000 [==============================] - 8s 192us/step - loss: 0.0593 - accuracy: 0.9830 - val_loss: 1.3019 - val_accuracy: 0.7263\n",
      "Epoch 79/100\n",
      "42000/42000 [==============================] - 7s 160us/step - loss: 0.0591 - accuracy: 0.9825 - val_loss: 1.2033 - val_accuracy: 0.7386\n",
      "Epoch 80/100\n",
      "42000/42000 [==============================] - 6s 139us/step - loss: 0.0573 - accuracy: 0.9828 - val_loss: 1.1879 - val_accuracy: 0.7564\n",
      "Epoch 81/100\n",
      "42000/42000 [==============================] - 6s 141us/step - loss: 0.0527 - accuracy: 0.9840 - val_loss: 1.3075 - val_accuracy: 0.7346\n",
      "Epoch 82/100\n",
      "42000/42000 [==============================] - 6s 138us/step - loss: 0.0514 - accuracy: 0.9849 - val_loss: 1.1502 - val_accuracy: 0.7581\n",
      "Epoch 83/100\n",
      "42000/42000 [==============================] - 6s 148us/step - loss: 0.0549 - accuracy: 0.9838 - val_loss: 1.3613 - val_accuracy: 0.7204\n",
      "Epoch 84/100\n",
      "42000/42000 [==============================] - 6s 150us/step - loss: 0.0540 - accuracy: 0.9838 - val_loss: 1.3583 - val_accuracy: 0.7217\n",
      "Epoch 85/100\n",
      "42000/42000 [==============================] - 6s 141us/step - loss: 0.0491 - accuracy: 0.9856 - val_loss: 1.2615 - val_accuracy: 0.7378\n",
      "Epoch 86/100\n",
      "42000/42000 [==============================] - 8s 187us/step - loss: 0.0482 - accuracy: 0.9863 - val_loss: 1.2808 - val_accuracy: 0.7311\n",
      "Epoch 87/100\n",
      "42000/42000 [==============================] - 7s 159us/step - loss: 0.0462 - accuracy: 0.9864 - val_loss: 1.2066 - val_accuracy: 0.7548\n",
      "Epoch 88/100\n",
      "42000/42000 [==============================] - 6s 143us/step - loss: 0.0509 - accuracy: 0.9850 - val_loss: 1.1714 - val_accuracy: 0.7535\n",
      "Epoch 89/100\n",
      "42000/42000 [==============================] - 6s 141us/step - loss: 0.0455 - accuracy: 0.9863 - val_loss: 1.3543 - val_accuracy: 0.7282\n",
      "Epoch 90/100\n",
      "42000/42000 [==============================] - 6s 154us/step - loss: 0.0521 - accuracy: 0.9845 - val_loss: 1.6254 - val_accuracy: 0.6912\n",
      "Epoch 91/100\n",
      "42000/42000 [==============================] - 6s 152us/step - loss: 0.0563 - accuracy: 0.9825 - val_loss: 1.5593 - val_accuracy: 0.7040\n",
      "Epoch 92/100\n",
      "42000/42000 [==============================] - 8s 193us/step - loss: 0.0504 - accuracy: 0.9846 - val_loss: 1.5926 - val_accuracy: 0.6937\n",
      "Epoch 93/100\n",
      "42000/42000 [==============================] - 8s 193us/step - loss: 0.0517 - accuracy: 0.9840 - val_loss: 1.2742 - val_accuracy: 0.7505\n",
      "Epoch 94/100\n",
      "42000/42000 [==============================] - 8s 182us/step - loss: 0.0455 - accuracy: 0.9865 - val_loss: 1.3548 - val_accuracy: 0.7288\n",
      "Epoch 95/100\n",
      "42000/42000 [==============================] - 7s 158us/step - loss: 0.0471 - accuracy: 0.9860 - val_loss: 1.3292 - val_accuracy: 0.7499\n",
      "Epoch 96/100\n",
      "42000/42000 [==============================] - 6s 150us/step - loss: 0.0426 - accuracy: 0.9875 - val_loss: 1.5047 - val_accuracy: 0.7199\n",
      "Epoch 97/100\n",
      "42000/42000 [==============================] - 6s 147us/step - loss: 0.0482 - accuracy: 0.9855 - val_loss: 1.5225 - val_accuracy: 0.7199\n",
      "Epoch 98/100\n",
      "42000/42000 [==============================] - 8s 202us/step - loss: 0.0433 - accuracy: 0.9865 - val_loss: 1.6567 - val_accuracy: 0.7045\n",
      "Epoch 99/100\n",
      "42000/42000 [==============================] - 9s 204us/step - loss: 0.0402 - accuracy: 0.9878 - val_loss: 1.2562 - val_accuracy: 0.7528\n",
      "Epoch 100/100\n",
      "42000/42000 [==============================] - 8s 202us/step - loss: 0.0448 - accuracy: 0.9864 - val_loss: 1.2986 - val_accuracy: 0.7474\n"
     ]
    }
   ],
   "source": [
    "# Fitting the ANN to the Training data\n",
    "history_final = keras_model_final.fit(X_train, y_train, validation_data=(X_test,y_test),batch_size = 200, epochs = 100, verbose = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "18000/18000 [==============================] - 2s 129us/step\n",
      "Test accuracy :  0.7473888993263245\n"
     ]
    }
   ],
   "source": [
    "results_final = keras_model_final.evaluate(X_test, y_test)\n",
    "print('Test accuracy : ', results_final[1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Predicting the digits using NN classifier on test data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Image 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1f1c9287240>"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "#Showing the image\n",
    "plt.imshow(X_test[5].reshape(32,32),cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Predicting the digits\n",
    "keras_model_final.predict_classes(X_test)[5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Image 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1f1c92de780>"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD5CAYAAADhukOtAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAXq0lEQVR4nO2dW4xcVXaG/2XjC223L+3GdrvtwR6DBZa5qkFIRCMSkpGDRgIeQMPDyA9oPA+DFKTJAyJSIG8kCox4QEgmWOOJCAMKV0UoGYQSoZECoTEGm9hkDDZ2t9tu20370rbxpVce6lhpmLP+qt7Vdcph/59kdfVevevss8/5XVX7r7W2uTuEEN99prV7AEKIapDYhcgEiV2ITJDYhcgEiV2ITJDYhciEy5rpbGbrATwNYDqAf3D3J9jfd3V1eW9vb/Rckz7+tGlp/1dduHAhjKVYkWwc06dPD2PsnMfHx8MYG/+5c+dK28+ePZv0fOfPnw9j7Nxmzpw56T4sxuaYzWN0buw6p8YYbIzRubHrcvLkydL20dFRjI2NlR4sWexmNh3AMwD+DMAAgA/M7E13/++oT29vL15//fXSGLvQl11WPsxZs2aFfdhFGRsbC2Nff/11GIvo6OgIY/Pnzw9j7AaORAsAX331VRgbGhoqbR8YGAj7jIyMhLEjR46Esa6urjC2bNmy0vZFixaFfdhcsWsd3R9ALAr2nx+bexZjgmZjnD17dmn78ePHwz7vvfdeafszzzwT9mnmbfytAHa7+xfufhbAbwDc3cTzCSFaSDNi7wWwf8LvA0WbEOISpBmxl71n+YP3zma20cz6zayfvV0UQrSWZsQ+AGDFhN+XAzjw7T9y903u3ufufewznhCitTQj9g8AXG1mq8xsJoAfA3hzaoYlhJhqklfj3f28mT0E4N9Qs942u/unrM/4+DhOnz5dGpsxY0Y8yGAlk624M+uKrYKzVd+U5ztz5kwYYyu7bCV2//79Yeyzzz4rbd+1a1fY5/Dhw2FseHg4jLGV9RUrVpS2X3XVVWGfVatWhbFodR/grkB0HzBbi8XYfcVgblN0PHZ/RDpimmjKZ3f3twC81cxzCCGqQd+gEyITJHYhMkFiFyITJHYhMkFiFyITmlqNn0qYzcCSFiJSM+KYtRLZcsw2ZDH2jcK9e/eGsR07doSxyGLbt29f2Ofo0aNhjFmHLEkmsuyYpcgsr9SMuCixKdVCYwktqVmM0f3NkrIi641azmFECPGdQmIXIhMkdiEyQWIXIhMkdiEyodLVeDMLa5OxVcRoRZh+6Z+smrK6amwcUYyNg5WQGhwcDGNsxX379u1h7ODBg6XtzO1gCS1sjMxNiGJs7hcsWBDGUtOjU2rhsXuHwe6DlDp/zCUZHR2d9HH0yi5EJkjsQmSCxC5EJkjsQmSCxC5EJkjsQmRC5YkwkX2VUjMutVYYizHbJYodO3Ys7PPBBx+EMVYXLjVxZe7cuaXtPT09YZ+Ucwa4jRbVyWOW1549e8IYswfZGJcuXRrGIpiFllKjEOBJMlHCC0vwmTNnzqT76JVdiEyQ2IXIBIldiEyQ2IXIBIldiEyQ2IXIhKasNzPbC+AEgAsAzrt7X70+kTXA6sxdfvnlpe3MemP2CYsx6yKqJ8eyv7Zu3RrGdu/eHcaYrcXGGFlUK1euDPuwuWfHGhsbC2ORPciy75ilyOZ4+fLlYSyyvFK3f2IWWnSfAtxyjO7HefPmhX0iS5FuoxZGGueP3T2uPCiEuCTQ23ghMqFZsTuA35rZh2a2cSoGJIRoDc2+jb/d3Q+Y2WIAb5vZLnd/d+IfFP8JbAT4trtCiNbS1Cu7ux8ofg4DeA3ArSV/s8nd+9y9b+HChc0cTgjRBMliN7M5ZtZ58TGAHwKIC6cJIdpKM2/jlwB4rbAiLgPwT+7+r6zDtGnTwqwhZk1EpFokzHpj1tDJkydL21lRyZ07d4Yxtn1SZ2dnGOvu7g5jkV0TZcMBfJshFmPv1ObPn1/azixFdl0YLBMtsqJSz7kVll2UtccKcEZWalRgE2hC7O7+BYAbUvsLIapF1psQmSCxC5EJErsQmSCxC5EJErsQmVBpwclp06aFmUEs8yqyw1iGD7PQUmPRGKN9twBeOJJZNcwqY7ElS5aUtjObjFlNrDgns4Y6OjpK25lNGe3pB/DrkpKJxmw+Nh/sPp09e3YYY5ZYNH72fNF1YZrQK7sQmSCxC5EJErsQmSCxC5EJErsQmVDparyZhauFKUkQbDWbwVZU2YpwtALKElpYDbeU5AiAr9JGq8+sPhqLsXGwhJzIMTh16lTSsdg5M3ci2iaJwcbIricbf0qyDk1qIceK0Cu7EJkgsQuRCRK7EJkgsQuRCRK7EJkgsQuRCZVabwxmJbBkjAiWOMEsu9OnT086xhJhmC2UagGyWGQNMeuHWW9R3T2AW1RRjNUa7OrqCmNXXnllUiw6bzaHbD7YvcPOjcWia8buxeHh4dJ2um1YGBFCfKeQ2IXIBIldiEyQ2IXIBIldiEyQ2IXIhLrWm5ltBvAjAMPuvq5o6wLwEoCVAPYCuN/d4+JiBe4eWmIs6y2yLVK320m13o4fPz7pPswKYfYPs+VOnDgRxqL5ZdbmokWLwlhk8QD83KIxpp4zs/lYzbgoc4yNncVasa1YxP79+8PYRx99VNpOM/YaOOavAKz/VtsjAN5x96sBvFP8LoS4hKkr9mK/9ZFvNd8NYEvxeAuAe6Z4XEKIKSb1M/sSdx8CgOLn4qkbkhCiFbR8gc7MNppZv5n1Hz16tNWHE0IEpIr9kJn1AEDxM1zFcfdN7t7n7n1sIUgI0VpSxf4mgA3F4w0A3pia4QghWkUj1tuLAO4A0G1mAwAeA/AEgJfN7EEA+wDc18jBzCwsHMgK+UWwTCJmNTF7gmVeRTYO23KHZeyxjDI2fvZxaM+ePaXtLPvu2LFjScdi8xjZV2NjY2GfoaGhMLZ169YwxrIOr7/++tJ2tnVV6jZOqduRRQVL2Tnv3r27tJ3ZkHXF7u4PBKE76/UVQlw66Bt0QmSCxC5EJkjsQmSCxC5EJkjsQmTCJbPXGyvyF9lXKftdAdyemDdvXhiLbKNdu3aFfZgdw7KrWAYVy4aKss0OHjwY9mHzyLLUBgcHw1g0x+ycmfXGMuIiG4o95y233BL2WbNmTRhj9ymzj5nNGt0/27ZtC/scPny4tJ1ZfHplFyITJHYhMkFiFyITJHYhMkFiFyITJHYhMqFS640VnGSF/CJrKNW6YploKf3mzJkT9lm2bFkYY+NnmWgscyzqx6wrVmeAFedk8xjNSeo1Y3Ypswcj621gYCDs093dHcaYvcbGyIqERtmD7LyiTDnt9SaEkNiFyAWJXYhMkNiFyASJXYhMqHw1PloVZquIUd0vtlJMEwIS6t0Bcc07Vrdu3bp1YayzszOMffnll2Hs888/D2PRFlWsXtzixXHZ/yuuuCKMsRqAUT+WxBONHUiv/Ra5E1EiCZC2nRTAE4pmzZoVxqKad+z+juaK9dEruxCZILELkQkSuxCZILELkQkSuxCZILELkQmNbP+0GcCPAAy7+7qi7XEAPwVw0b941N3fauSAUcILswwiu44lYrAYI6UfS5y44YYbwhjbkolZZSyJI0omYUkVzLpauXJlGGP12KKadyzBh9VpY9eFnVtkwbJtrU6fPh3GmG3LrDdms0bWLbN0Ozo6StvZ+Bp5Zf8VgPUl7b909xuLfw0JXQjRPuqK3d3fBTBSwViEEC2kmc/sD5nZJ2a22cwWTtmIhBAtIVXszwJYDeBGAEMAnoz+0Mw2mlm/mfWzz0lCiNaSJHZ3P+TuF9x9HMBzAG4lf7vJ3fvcvY9VRBFCtJYksZtZz4Rf7wWwY2qGI4RoFY1Yby8CuANAt5kNAHgMwB1mdiMAB7AXwM8aOZiZhfYEyzSKbBeWKcdirN4dq4MWZXmxdyzf+973whizavbu3RvG5s+fH8ai+nTM5mMWz/Lly8MYe85org4cOBD2YXYjs8NYv8iKGhmJ15yZPchsPmYPsnsusjDZdY7sXraFVl2xu/sDJc3P1+snhLi00DfohMgEiV2ITJDYhcgEiV2ITJDYhciESgtOmlloybCCfMx2iaDZPyTGrLcoxiwSdl4sxmBZatG2S/PmzQv7sFiUXQXw844yFXt6ekrbAW6Xsm9fsu2womvG7DVm5THrjVmpKfPIrsuCBQtK21kRUL2yC5EJErsQmSCxC5EJErsQmSCxC5EJErsQmVCp9QbEthcrOJlCajHKlEw6ZuWNjo6GMbbv2eDgYBhjNlRUYDE10y+y0OrFoj3RFi6Mixqxvd6YVcbGH10bZpMx2DwymCUW7fW2evXqsE90L27dujXso1d2ITJBYhciEyR2ITJBYhciEyR2ITKh8tX4aNWd1aCLVjJTt3hiK/9sNT5afWZjZyvnu3fvDmOHDh0KYydOnAhj0RgXL14c9mHzyFbc2Qp5lOTDtso6cuRIGGOOB0sMipJk2Gp85CTU68e2w2Kr+NH9za5ZdA+wseuVXYhMkNiFyASJXYhMkNiFyASJXYhMkNiFyIRGtn9aAeDXAJYCGAewyd2fNrMuAC8BWInaFlD3u/tX7LnGx8dDK4fZYZE1wfqwWKplF9knLBGDbcdz+PDhMMYsO5YwEs0Vs65YLTx2LLb9U1RXjY1jeHg4jDF7jc1/dK2ZhcbmI0paAbj1xiyxaPwp14yNr5FX9vMAfuHu1wK4DcDPzWwtgEcAvOPuVwN4p/hdCHGJUlfs7j7k7luLxycA7ATQC+BuAFuKP9sC4J5WDVII0TyT+sxuZisB3ATgfQBL3H0IqP2HACD+uo8Qou00LHYzmwvgFQAPu3v8Qe4P+200s34z62efQ4UQraUhsZvZDNSE/oK7v1o0HzKzniLeA6B0dcXdN7l7n7v3sX3MhRCtpa7YrbYE/TyAne7+1ITQmwA2FI83AHhj6ocnhJgqGsl6ux3ATwBsN7NtRdujAJ4A8LKZPQhgH4D7GjlgZDOk2GipteRYP2bjpHDy5MkwxmwtltnGtsOKLJmUrCuA20nRVlMsxq5zqnXFxhFZvcyiYvORUksO4POfug1YGWx8dcXu7r8DEI30zsQxCSEqRt+gEyITJHYhMkFiFyITJHYhMkFiFyITKi84mZI5dubMmdJ2lhWUuv0Ti7HjRbQiS4rZOJHlyIpiRvMLAB0dHWGMZb1F42fXOTWLkWWwdXZ2lrYvWLAg7MPsq5R7AODjj56TXbMoRrfCCiNCiO8UErsQmSCxC5EJErsQmSCxC5EJErsQmVCp9ebuoWXAbIaUrLfU7DVmrUSWDMuwYxYPi0UFGwHgq6/iup7nzp0rbT916lTYZ2RkJIyxophLliwJY9H8s2y+aF+2ev3YfnSrV68ubb/qqqvCPmzuGeyeS8nCZNlw0X3K7l+9sguRCRK7EJkgsQuRCRK7EJkgsQuRCZWuxl+4cAHHjh0rjbG6aiwpJCI1YSEluYYlpjBYAgdLoGHzESW1pK7GDw4OhrEoyQSIHRRWTnz//v1hbHR0NIyxeVy7dm1p+zXXXBP2YVWQ2TZULIGGuU1RQhS7ZilbmOmVXYhMkNiFyASJXYhMkNiFyASJXYhMkNiFyIS61puZrQDwawBLAYwD2OTuT5vZ4wB+CuBipsSj7v4Wey5mvbFEgRTrjVkTzKqZ6hp0zD5hsCQINh/RGNl5sfpozDJidlgUO3jwYNiHxZi9uXhxvFv4mjVrStuXLVsW9mG19dh1YdYbs+yi+zFlWyh2jzbis58H8At332pmnQA+NLO3i9gv3f3vJz0iIUTlNLLX2xCAoeLxCTPbCaC31QMTQkwtk3pfamYrAdwE4P2i6SEz+8TMNpvZwikemxBiCmlY7GY2F8ArAB529+MAngWwGsCNqL3yPxn022hm/WbWzz7jCSFaS0NiN7MZqAn9BXd/FQDc/ZC7X3D3cQDPAbi1rK+7b3L3PnfvY5VZhBCtpa7YrbYM+jyAne7+1IT2ngl/di+AHVM/PCHEVNHIavztAH4CYLuZbSvaHgXwgJndCMAB7AXws3pP5O6hxZa6lVMKbLsjNo6oxhirPcZizP6JLCOA25RRthx7V8Vqya1atSqMseywyB48efJk2Gfp0qVhjM3HtddeG8auu+660nZmk7EYm3tWJ4/ZvZG9yTIfo/uKWZSNrMb/DkDZM1BPXQhxaaFv0AmRCRK7EJkgsQuRCRK7EJkgsQuRCZUWnJw+fXpoAU11tlmqlcdika3B7A5ma0VbNQF8i6eU4ousOGSqLcest6iIIoNlCHZ1dYWxqKgkwG20CJo5Riy0lK3DWCz1Po3QK7sQmSCxC5EJErsQmSCxC5EJErsQmSCxC5EJlVtvkQXEbKizZ8+WtqdaEymZbUCaRcKKIbIMKrb3HTvevHnzSttZwcPu7u4wxuaKFaqMiiXOnz8/7MPsQWYB9vZOvkoaG3tKcch6/Zg9G2UIMk2kHEev7EJkgsQuRCZI7EJkgsQuRCZI7EJkgsQuRCZUar0BsZXDbIvIgmCZRMyCYKT0Y3YM25ctsskAXnyR2YNRkUJmNS1cGO/vwbLX2Lml7FPG5p4V52TZd1GGILMUWYzdp+x+TMmWS7EAZb0JISR2IXJBYhciEyR2ITJBYhciE+quxpvZbADvAphV/P0/u/tjZtYF4CUAK1Hb/ul+d48Lp/3f85W2s+SOlDpibMU6paYde062ApqSAAHwunBsPqLVc1bfjW0zxBJ52BxHq8Vs7OweYAk0zDFg5x3B7g92zaZ6NZ4lwqTci43c9V8D+BN3vwG17ZnXm9ltAB4B8I67Xw3gneJ3IcQlSl2xe42Lu/HNKP45gLsBbCnatwC4pyUjFEJMCY3uzz692MF1GMDb7v4+gCXuPgQAxc/4/Z4Qou00JHZ3v+DuNwJYDuBWM1vX6AHMbKOZ9ZtZ/8jISOo4hRBNMqmVKncfBfAfANYDOGRmPQBQ/BwO+mxy9z5372OF/oUQraWu2M3sCjNbUDy+HMCfAtgF4E0AG4o/2wDgjVYNUgjRPI0kwvQA2GJm01H7z+Fld/8XM/tPAC+b2YMA9gG4r94TmRm1ICIiS4ZZP/XGEZFi2VG7g9g4zPJiNk5HR0cYi6yhsbGxsA+7JnPmzAljzBpKSShiyTqsXh87txSbNWWrpnrHYkktEeyckxK26v2Bu38C4KaS9qMA7pz0EYUQbUHfoBMiEyR2ITJBYhciEyR2ITJBYhciEyzVvko6mNlhAF8Wv3YDOFLZwWM0jm+icXyT/2/juNLdrygLVCr2bxzYrN/d+9pycI1D48hwHHobL0QmSOxCZEI7xb6pjceeiMbxTTSOb/KdGUfbPrMLIapFb+OFyIS2iN3M1pvZZ2a228zaVrvOzPaa2XYz22Zm/RUed7OZDZvZjgltXWb2tpn9vvgZ78nU2nE8bmaDxZxsM7O7KhjHCjP7dzPbaWafmtlfFO2VzgkZR6VzYmazzey/zOzjYhx/U7Q3Nx/uXuk/ANMBfA7g+wBmAvgYwNqqx1GMZS+A7jYc9wcAbgawY0Lb3wF4pHj8CIC/bdM4HgfwlxXPRw+Am4vHnQD+B8DaqueEjKPSOQFgAOYWj2cAeB/Abc3ORzte2W8FsNvdv3D3swB+g1rxymxw93cBfLtGV+UFPINxVI67D7n71uLxCQA7AfSi4jkh46gUrzHlRV7bIfZeAPsn/D6ANkxogQP4rZl9aGYb2zSGi1xKBTwfMrNPirf5Lf84MREzW4la/YS2FjX91jiAiuekFUVe2yH2shIb7bIEbnf3mwH8OYCfm9kP2jSOS4lnAaxGbY+AIQBPVnVgM5sL4BUAD7v78aqO28A4Kp8Tb6LIa0Q7xD4AYMWE35cDONCGccDdDxQ/hwG8htpHjHbRUAHPVuPuh4obbRzAc6hoTsxsBmoCe8HdXy2aK5+TsnG0a06KY0+6yGtEO8T+AYCrzWyVmc0E8GPUildWipnNMbPOi48B/BDADt6rpVwSBTwv3kwF96KCObFaQbXnAex096cmhCqdk2gcVc9Jy4q8VrXC+K3VxrtQW+n8HMBftWkM30fNCfgYwKdVjgPAi6i9HTyH2judBwEsQm0brd8XP7vaNI5/BLAdwCfFzdVTwTj+CLWPcp8A2Fb8u6vqOSHjqHROAFwP4KPieDsA/HXR3tR86Bt0QmSCvkEnRCZI7EJkgsQuRCZI7EJkgsQuRCZI7EJkgsQuRCZI7EJkwv8Cd7MvME0ICzgAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Showing the image\n",
    "plt.imshow(X_test[10].reshape(32,32),cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Predicting the digits\n",
    "keras_model_final.predict_classes(X_test)[10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Image 3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1f1c9345358>"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD5CAYAAADhukOtAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAXBUlEQVR4nO2dX2xdVXbGv2XHDvnrxI6TOH8gIQKpgDoJsiIE1YiWdpSikYAH0PAwygOazMMgFWn6gEAq9I1WhREPFVIo0WQqyoAKCFShdlA0FRqpohgKIWmmDIPcTBrHTsg/J4E4iVcf7klrwlnfvd733nNN9veTIttn3X3PuvucFd+7P39rm7tDCHH109XpBIQQ1aBiFyITVOxCZIKKXYhMULELkQkqdiEyYV4zg81sG4BnAXQD+Dt3f4o9vre31xcsWFAaS5EAu7rS/q+anp4OY2YWxi5evJh0vgj2mlke8+bFl627u7upnGaTR8o1Y3N/6dKlpBgjmo+enp5wDHvN7J5LHRflmHKdjx49isnJydJEkovdzLoB/C2APwFwCMB7Zvamu/9nNGbBggW4/fbbS2MXLlwIzxVNVG9v7ywy/n+mpqbCGLtgx48fLz3OLiT7D4LdwOxmHBgYCGNLly4tPc4Kk91UqcUexb744otwTDS/AHDmzJmkPJYsWVJ6fM2aNeEYNvfXXHNNUmzRokVhrK+vr/R4ynV+/PHHwzHNvI3fCuBTd//M3acA/BzAPU08nxCijTRT7GsB/G7Gz4eKY0KIOUgzn9nL3t997f2Ume0AsAPgb3OEEO2lmd/shwCsn/HzOgCHr3yQu+9092F3H079jC2EaJ5miv09ADeY2UYz6wXwPQBvtiYtIUSrSX4b7+4XzexhAP+CmvS2y933szFmFq78RpIcEK+6MxmHrYKfO3cujJ06dWrWMbaqnroKvnz58jC2cOHCMNbf3196nL2rSpl7gKsakbrCVtXZPLJxZ8+eDWPnz58vPc5W3Nl1YR9F2XVh91wko7W6JprS2d39LQBvNfMcQohq0F/QCZEJKnYhMkHFLkQmqNiFyAQVuxCZ0NRq/KxPNm8eVqxYURpjUlMkdzB5jRkuIuMBAIyNjYWxSMZh0g8zkrA8hoaGwtj69evD2HXXXVd6nBkxmLzGTD4p0hszu7DrySRRJstFz3n69OlwDJPlWI5M9mLz+OWXX876+ZJcorMeIYT4RqJiFyITVOxCZIKKXYhMULELkQmVrsbPnz8fGzduLI1Fq/RAbAhgq5UMtlJ/+PDXXLr/R2TUOHLkSDiGmSrYivu6devC2IYNG8LY9ddfX3p88eLF4ZjUVd+UnnGRUaceTPGIVrOBeKWe5c4UFNY+LbUFWQS7dyLFgPbBm3UGQohvJCp2ITJBxS5EJqjYhcgEFbsQmaBiFyITKjfCDA4OlsaYnBSZZJjUwcwdzDjBTBCjo6Olx5mUx/JYvXp1GGOyHJMpI2mLyTiRwace7Dmj183MPyyPY8eOhbETJ06EsUg6TN1ei5FqhEkhylHSmxBCxS5ELqjYhcgEFbsQmaBiFyITVOxCZEJT0puZjQKYBHAJwEV3H2aP7+7uxpIlS0pjrEdatPE8cyCxrXhYjPVIi2TDkydPhmPYdkHMAcZ68rG5imD94pgUySQjtj1RlCMbs3LlyjDG5EY2V9E9wuaDvWYmy0XbOAHcPRjFWi7XteA5/tDdYxFUCDEn0Nt4ITKh2WJ3AL8ws/fNbEcrEhJCtIdm38bf4e6HzWwlgLfN7Nfu/s7MBxT/CewAgIGBgSZPJ4RIpanf7O5+uPg6AeB1AFtLHrPT3YfdfThaaBNCtJ/kYjezRWa25PL3AL4DYF+rEhNCtJZm3savAvB64bKZB+Af3P2f2YCurq5QRmMyw/j4eOlx1oRw1apVYWzZsmUtHcekMCa9RTJkvXFsriLnFWvKyGLsXExqivJgrqz58+eHMSaXMjkvcjEy6Y3JZMxpyWIp52MuuqQGlrMeUeDunwH4Vup4IUS1SHoTIhNU7EJkgopdiExQsQuRCSp2ITKh0oaT09PToVy2d+/eWT8fc0IxWYs1emSxaL80di7mkmJurZtvvjmMMbfc6dOnS49H8iUA7N+/P4yxZppbtmwJY9Gec0ymZFIT++tLJr1Fkh07V8rzAfw+YLFI0mXSZoojTr/ZhcgEFbsQmaBiFyITVOxCZIKKXYhMqHQ1nsH6oEUrp8zQwmBGB0a0SssMHAy21RTbNoqNiwwSzIjBVtyZSSYFtoqcarphikd0zdi5mBWbxSK1BkhbqWfniu4Bbf8khFCxC5ELKnYhMkHFLkQmqNiFyAQVuxCZMGekNyoZBDJJqlGAyVpMaoqkFSa5sL5kKRJaveeMYsz4waQrZlxh4yJYHqmSKJvHqJcfM7swSZeZkFKkMiA217Aco9dFpc0wIoS4qlCxC5EJKnYhMkHFLkQmqNiFyAQVuxCZUFc7MbNdAL4LYMLdbymO9QN4GcAGAKMAHnD3E/Wey91DaYjJSZGcwGSGFNkC4O6waBx7PiahpTrbmEwZnY8529hrZvJaikuNPR9zD6b2d4vuq9TegKwXHpPs2Fyx6xkRvS4mXzbym/2nALZdcexRAHvc/QYAe4qfhRBzmLrFXuy3fvyKw/cA2F18vxvAvS3OSwjRYlI/s69y9zEAKL6ubF1KQoh20PYFOjPbYWYjZjYyOTnZ7tMJIQJSi33czIYAoPg6ET3Q3Xe6+7C7D7OFFCFEe0kt9jcBbC++3w7gjdakI4RoF41Iby8BuBPACjM7BOAJAE8BeMXMHgJwEMD9jZzM3UOZJ0XyYtIVk8NSpbJITkptosjkQRZjslx0PuY2u3DhQlKMyaWRtMVeF7sH2LtC5jo8f/586XG2jROT3liM5cGIrg27F6MxTHqrW+zu/mAQuqveWCHE3EF/QSdEJqjYhcgEFbsQmaBiFyITVOxCZEKlDSeZ641JPJEbKqXRYL0Yk5MidxKTSFLdTsyVxWKRHMleM5MwU+aDxdiY1GvG5LyoCSRrHNnX1xfGmLON5cEksRQ5OrrntNebEELFLkQuqNiFyAQVuxCZoGIXIhNU7EJkQqXSm5mFshHbty0FJnkxBxiTmqJxTO5gz8f2lWNSJJOhIjcXk5pYLMWNCMRzxZ4vRWoC0hpVMocak9DYudg9zPKPZLnUve8i9JtdiExQsQuRCSp2ITJBxS5EJqjYhciESlfju7q66GpmRLSinbqqzlZG2TZJKT3o2POdO3cuaRzrxxaZWiJDCACsXBm3/Wd5pBg/mAJx5syZMMbUidQtpVJoRy+/qE9edJzl0ez2T0KIqwAVuxCZoGIXIhNU7EJkgopdiExQsQuRCY1s/7QLwHcBTLj7LcWxJwH8AMDR4mGPuftb9Z6rp6cHa9euLY3deOONcZKBtLJmzRp6roizZ8+GMbbTLDO8RDBzB8uDxZh0uGjRotLjg4OD4RgmCzHpjcl5EamSaGq/vug5Wd89JommyGEAvw8iOTJl67BmpbefAthWcvwn7r65+Fe30IUQnaVusbv7OwCOV5CLEKKNNPOZ/WEz22tmu8ws3tpSCDEnSC325wBsArAZwBiAp6MHmtkOMxsxs5FTp04lnk4I0SxJxe7u4+5+yd2nATwPYCt57E53H3b3YdZ8XwjRXpKK3cyGZvx4H4B9rUlHCNEuGpHeXgJwJ4AVZnYIwBMA7jSzzQAcwCiAHzZysp6eHiqXRUTOJfZcbJseJtWwPmKRFMIkF+bkOn48XvdkMSajRa+bOeVWr14dxpgcxqS3SBpKkcnqjUvdviqCSW8sRybLsViUY9RPEIj7/zHprW6xu/uDJYdfqDdOCDG30F/QCZEJKnYhMkHFLkQmqNiFyAQVuxCZUPn2T5F8xbYSirbqYVv4MNkixU0ExLIca6J48uTJMMZgWzKx1x3JOKzxInOUsesSOeyAeP6ZNJTiKgS4LBe5H5mbjznUmITG3IMpW1sxSTG659RwUgihYhciF1TsQmSCil2ITFCxC5EJKnYhMqFS6e3ixYs4duxYaYw5jSL5h7nNUveBY+MimBxz9OjRMMYccamyYpQ/c72x18zkHybLRePYXJ04cSKMsfsj5Zox9xp7PjaO3VcpsZT7VNKbEELFLkQuqNiFyAQVuxCZoGIXIhMqXY2fmprCwYMHS2PMRBAZJFifOWb8YH3m2ApotNKZ2oOObTXFTCas91tk8mFzxUwabK6YmSQyarDV7NOnT4cxNo/smkX5s+3BmMrAYHPFYtH8szGR2sHMRPrNLkQmqNiFyAQVuxCZoGIXIhNU7EJkgopdiExoZPun9QB+BmA1gGkAO939WTPrB/AygA2obQH1gLvHTgbUJIaxsbHSGDNIRCxfHu8UzaQrJq0wI0GKYYHBzsVgcxX1vGMy2dmzZ8MY69fHJMdISmWSV8o9APAedFH+TJpl5h92XzGYkSe6NizHqFciu16N/Ga/CODH7v57AG4D8CMzuwnAowD2uPsNAPYUPwsh5ih1i93dx9z9g+L7SQAHAKwFcA+A3cXDdgO4t11JCiGaZ1af2c1sA4AtAN4FsMrdx4DafwgAVrY6OSFE62i42M1sMYBXATzi7vHfNX593A4zGzGzEfbZUAjRXhoqdjPrQa3QX3T314rD42Y2VMSHAEyUjXX3ne4+7O7DqYsbQojmqVvsVvvL+hcAHHD3Z2aE3gSwvfh+O4A3Wp+eEKJVNOJ6uwPA9wF8bGYfFsceA/AUgFfM7CEABwHcX++JWA+66DgQO6j6+vrCMexdBOvHxj5qRPIJk95YHiw2NDQUxtjWUCnbHZ06dSqMMXmQSVQRrLcek42YvMYkwMgFxnJnMeZEYzmy+U+5r6Ic6RyGkQJ3/xWAyDd3V73xQoi5gf6CTohMULELkQkqdiEyQcUuRCao2IXIhEobTppZKF2wZoORw4c5iVKdaGwca3oYwWS+devWhbFNmzaFsVWrVs06j8gNB/AtqtgcR9cFiB1brCFiqhuRNbFMORdz5jG5lLnU2L0TzUkkOQNcYgvHzHqEEOIbiYpdiExQsQuRCSp2ITJBxS5EJqjYhciESqW3efPmYXBwsDQ2MVFqhwcQO42Y5EL3vCKyBXM8RVIIa5TI5Bi2/xqT19auXRvGojlhkiKbD7YPXIpcyiQvds1YHkzWimS0hQsXhmPYNWOyXMrebOw5U+XjCP1mFyITVOxCZIKKXYhMULELkQkqdiEyodLV+N7e3tD8MT4+Ho6L+sIxA8fnn38expg5Jdq2CIhXn9m5GMzowPqZRYoGEK+CsxVrtlLM+rux3nXRHLOVf7b6PDo6GsZY/8Jrr7229PjAwEA4ZunSpWGMGXKYaYit1EfKQIr5h6pQYUQIcVWhYhciE1TsQmSCil2ITFCxC5EJKnYhMqGu9GZm6wH8DMBqANMAdrr7s2b2JIAfALisfz3m7m+x53L3UF5hUlNknmAyDpOMmKmCbdMTyR3sXEw+YXkwWY5tURXlwuaKyY3sXMwwEl1n9rqYyYTNI5ObomuW0sOtmXGMKH/2mlNoRGe/CODH7v6BmS0B8L6ZvV3EfuLuf9PSjIQQbaGRvd7GAIwV30+a2QEAscdSCDEnmdV7EjPbAGALgHeLQw+b2V4z22Vmy1ucmxCihTRc7Ga2GMCrAB5x99MAngOwCcBm1H7zPx2M22FmI2Y2wpodCCHaS0PFbmY9qBX6i+7+GgC4+7i7X3L3aQDPA9haNtbdd7r7sLsPs785FkK0l7rFbrWlwhcAHHD3Z2YcH5rxsPsA7Gt9ekKIVtHIavwdAL4P4GMz+7A49hiAB81sMwAHMArgh80kwnqCRXIHc2sxB9Xk5GQYO3HiRBiLXHbs+ZhUw8YxJx3byimSa5iMw7ZxYu/GmJMret1MbmQx5rBj1yzarim1RyEbx+RjBnvOVtLIavyvAJRlQzV1IcTcQn9BJ0QmqNiFyAQVuxCZoGIXIhNU7EJkQqUNJ909dFgxGSeSf5hEwtxajCNHjoSxSA5jshCTVVjjSyavsVjkHGPz29/fH8aYhMlcaosXL551Hkx6Y3mwcZELkDkVU6W3VGdeBJsrln+EfrMLkQkqdiEyQcUuRCao2IXIBBW7EJmgYhciEyqV3swslBMidxIQu4lYw0MmTTBZju0bdvz48dLj58+fD8cwJ1Sq+45JfcuWLSs9zuaKSW+pRLJc1AAS4BIak7VS3HcpY+qNY68txWWX+prD88x6hBDiG4mKXYhMULELkQkqdiEyQcUuRCao2IXIhEqlt66urtDpxRxgkcyQuhcWk8pYLJKGWB7MrXXmzJkwxhpOjo2NhbFoX7y+vr5wDJM9mWTH9sU7d+5c6XEmG6Y6/ZiUGu0tlyqhsRiDSW9RLCVHKvGFESHEVYWKXYhMULELkQkqdiEyQcUuRCbUXY03s2sAvANgfvH4f3T3J8ysH8DLADagtv3TA+4euzdQM6dEq7Rs1TdalYz62QF8pZit7B4+fDiMRUaY1H53bBzb2oqZfKIV/oGBgXDMihUrwhi7LilqAjMasV1+o/umHtG9k2p2YTGmyrDzVUUjGZwH8Efu/i3UtmfeZma3AXgUwB53vwHAnuJnIcQcpW6xe43L/4X3FP8cwD0AdhfHdwO4ty0ZCiFaQqP7s3cXO7hOAHjb3d8FsMrdxwCg+LqyfWkKIZqloWJ390vuvhnAOgBbzeyWRk9gZjvMbMTMRthnMiFEe5nVqoG7nwTwrwC2ARg3syEAKL5OBGN2uvuwuw+zvb6FEO2lbrGb2aCZLSu+XwDgjwH8GsCbALYXD9sO4I12JSmEaJ5GjDBDAHabWTdq/zm84u7/ZGb/BuAVM3sIwEEA99d7oqmpKYyOjpbGou2CgNjcweQYJgtNTJS+CQEAfPLJJ2Eskt5YLzkmD7Zju6PIyMPkRma6YT30mGkokhXZXDG5kZl1ovsDAAYHB0uPs+uSKpOlbv+UsjVUCnWL3d33AthScvxzAHe1IykhROvpvNIvhKgEFbsQmaBiFyITVOxCZIKKXYhMsNQ+bkknMzsK4L+LH1cAiC1Q1aE8vory+CrftDyuc/dSvbHSYv/Kic1G3H24IydXHsojwzz0Nl6ITFCxC5EJnSz2nR0890yUx1dRHl/lqsmjY5/ZhRDVorfxQmRCR4rdzLaZ2X+Z2adm1rHedWY2amYfm9mHZjZS4Xl3mdmEme2bcazfzN42s98UX5d3KI8nzex/ijn50MzuriCP9Wb2SzM7YGb7zezPiuOVzgnJo9I5MbNrzOzfzeyjIo+/LI43Nx/uXuk/AN0AfgvgegC9AD4CcFPVeRS5jAJY0YHzfhvArQD2zTj21wAeLb5/FMBfdSiPJwH8ecXzMQTg1uL7JQA+AXBT1XNC8qh0TgAYgMXF9z0A3gVwW7Pz0Ynf7FsBfOrun7n7FICfo9a8Mhvc/R0AV5rjK2/gGeRROe4+5u4fFN9PAjgAYC0qnhOSR6V4jZY3ee1Esa8F8LsZPx9CBya0wAH8wszeN7MdHcrhMnOpgefDZra3eJvf9o8TMzGzDaj1T+hoU9Mr8gAqnpN2NHntRLGXteXolCRwh7vfCuBPAfzIzL7doTzmEs8B2ITaHgFjAJ6u6sRmthjAqwAecfeOdSctyaPyOfEmmrxGdKLYDwFYP+PndQDibVjaiLsfLr5OAHgdtY8YnaKhBp7txt3HixttGsDzqGhOzKwHtQJ70d1fKw5XPidleXRqTopzz7rJa0Qniv09ADeY2UYz6wXwPdSaV1aKmS0ysyWXvwfwHQD7+Ki2MicaeF6+mQruQwVzYrUmbC8AOODuz8wIVTonUR5Vz0nbmrxWtcJ4xWrj3aitdP4WwOMdyuF61JSAjwDsrzIPAC+h9nbwAmrvdB4CMIDaNlq/Kb72dyiPvwfwMYC9xc01VEEef4DaR7m9AD4s/t1d9ZyQPCqdEwC/D+A/ivPtA/AXxfGm5kN/QSdEJugv6ITIBBW7EJmgYhciE1TsQmSCil2ITFCxC5EJKnYhMkHFLkQm/C/zVJ4GvaggMwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Showing the image\n",
    "plt.imshow(X_test[112].reshape(32,32),cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "9"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Predicting the digits\n",
    "keras_model_final.predict_classes(X_test)[112]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[6 7 2 ... 7 5 2]\n"
     ]
    }
   ],
   "source": [
    "#Predicting for all images\n",
    "y_pred=keras_model_final.predict_classes(X_test)\n",
    "print(y_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Differences and trade-offs between traditional KNN and NN classifiers with the help of classification metrics "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* KNN requires more time to find the nearest neibhours. It takes less time to train the model. It took more than 24hrs to find the best neighbour(k). However, Neural Network takes more time to train the model.\n",
    "* In KNN, we have only one hyperparameter k while in NN we have so many hyperparameters(Hidden Layers, Hidden Units,learning Rate, Number of Epochs, activations Functions).\n",
    "* KNN classification metrics shows that the accuracy is 46%. The digit 4 and 6 are classified correctly and achieved approx 60% accuracy. the other digits are not predicted correclty.\n",
    "* Using NN with batchNormalization we got 74.7% accuracy. We have predicted the classification of the images correctly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "name": "Project-8  Image classification in SVHN Dataset-Copy1.ipynb",
   "provenance": []
  },
  "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": 1
}