baseline_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "f1bbb1e1",
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5",
    "papermill": {
     "duration": 0.016411,
     "end_time": "2022-05-09T17:46:23.781761",
     "exception": false,
     "start_time": "2022-05-09T17:46:23.765350",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# Baseline Model\n",
    "- We will be using the processed CIFAR-10 dataset to train a baseline CNN model.\n",
    "- In the baseline model, we will be using the labelled dataset only, and will be testing the model on the test dataset.\n",
    "- I have also tried a little bit of hyper-parameter tuning in this kernel, regarding the architecture of CNN model, #epochs, etc.\n",
    "- The hyper-parameter tuning regarding the **#epochs** is fixed and carried out similarly in all the kernels. We have tried **[10, 20, 30, 40, 50]** epochs, and for the final model, selected the setting (#epochs) with the largest test accuracy. \n",
    "- There is one slight catch with this kernel. With the GPU, it is not possible to produce 100% reproducible results, even after setting the seeds of all the libraries used. For reference, check this [kernel](https://www.kaggle.com/lbronchal/keras-gpu-cpu-reproducibility-test) out."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "78d9dbbd",
   "metadata": {
    "papermill": {
     "duration": 0.015058,
     "end_time": "2022-05-09T17:46:23.813554",
     "exception": false,
     "start_time": "2022-05-09T17:46:23.798496",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 1. Importing the Packages & Boilerplate Code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "8c315a85",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:23.854857Z",
     "iopub.status.busy": "2022-05-09T17:46:23.854088Z",
     "iopub.status.idle": "2022-05-09T17:46:29.751331Z",
     "shell.execute_reply": "2022-05-09T17:46:29.750313Z",
     "shell.execute_reply.started": "2022-05-09T17:44:54.108420Z"
    },
    "papermill": {
     "duration": 5.922907,
     "end_time": "2022-05-09T17:46:29.751502",
     "exception": false,
     "start_time": "2022-05-09T17:46:23.828595",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import random\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from tqdm import tqdm\n",
    "from tabulate import tabulate\n",
    "from sklearn.metrics import accuracy_score, log_loss, f1_score\n",
    "\n",
    "# https://www.kaggle.com/c/ventilator-pressure-prediction/discussion/274717\n",
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' \n",
    "\n",
    "import tensorflow as tf\n",
    "import tensorflow.keras.layers as tfl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "6f7c768f",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:29.786257Z",
     "iopub.status.busy": "2022-05-09T17:46:29.785441Z",
     "iopub.status.idle": "2022-05-09T17:46:29.787760Z",
     "shell.execute_reply": "2022-05-09T17:46:29.787367Z",
     "shell.execute_reply.started": "2022-05-09T17:44:56.020643Z"
    },
    "papermill": {
     "duration": 0.021153,
     "end_time": "2022-05-09T17:46:29.787864",
     "exception": false,
     "start_time": "2022-05-09T17:46:29.766711",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Setting the seeds\n",
    "SEED = 0\n",
    "os.environ['PYTHONHASHSEED']=str(SEED)\n",
    "random.seed(SEED)\n",
    "np.random.seed(SEED)\n",
    "tf.random.set_seed(SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "516c41a1",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:32.236687Z",
     "iopub.status.busy": "2022-05-09T17:46:32.235877Z",
     "iopub.status.idle": "2022-05-09T17:46:32.239153Z",
     "shell.execute_reply": "2022-05-09T17:46:32.239557Z",
     "shell.execute_reply.started": "2022-05-09T17:44:56.027323Z"
    },
    "papermill": {
     "duration": 2.437134,
     "end_time": "2022-05-09T17:46:32.239695",
     "exception": false,
     "start_time": "2022-05-09T17:46:29.802561",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found GPU at: /device:GPU:0\n"
     ]
    }
   ],
   "source": [
    "# Making sure that Tensorflow is able to detect the GPU\n",
    "device_name = tf.test.gpu_device_name()\n",
    "if \"GPU\" not in device_name:\n",
    "    print(\"GPU device not found\")\n",
    "print('Found GPU at: {}'.format(device_name))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "9d6ede2a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:32.278654Z",
     "iopub.status.busy": "2022-05-09T17:46:32.277108Z",
     "iopub.status.idle": "2022-05-09T17:46:32.279279Z",
     "shell.execute_reply": "2022-05-09T17:46:32.279685Z",
     "shell.execute_reply.started": "2022-05-09T17:44:56.692773Z"
    },
    "papermill": {
     "duration": 0.024584,
     "end_time": "2022-05-09T17:46:32.279808",
     "exception": false,
     "start_time": "2022-05-09T17:46:32.255224",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# These are the usual ipython objects\n",
    "ipython_vars = ['In', 'Out', 'exit', 'quit', 'get_ipython', 'ipython_vars']\n",
    "\n",
    "# Defining a function to list the memory consumed\n",
    "# Only outputs variables taking at least 1MB space\n",
    "def list_storage(inp_dir):\n",
    "    # Get a sorted list of the objects and their sizes\n",
    "    vars_defined = [x for x in inp_dir if not x.startswith('_') and x not in sys.modules and x not in ipython_vars]\n",
    "    sto = sorted([(x, sys.getsizeof(globals().get(x))) for x in vars_defined], key=lambda x: x[1], reverse=True)\n",
    "    sto = [(x[0], str(round((x[1] / 2**20), 2)) + ' MB') for x in sto if x[1] >= 2**20]\n",
    "    print(tabulate(sto, headers = ['Variable', 'Storage (in MB)']))\n",
    "\n",
    "# In order to use this function, use the below line of code\n",
    "# list_storage(dir())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1497415c",
   "metadata": {
    "papermill": {
     "duration": 0.015007,
     "end_time": "2022-05-09T17:46:32.309877",
     "exception": false,
     "start_time": "2022-05-09T17:46:32.294870",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 2. Data Exploration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "10be32e9",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:32.347211Z",
     "iopub.status.busy": "2022-05-09T17:46:32.346702Z",
     "iopub.status.idle": "2022-05-09T17:46:54.411341Z",
     "shell.execute_reply": "2022-05-09T17:46:54.411819Z",
     "shell.execute_reply.started": "2022-05-09T17:44:56.701839Z"
    },
    "papermill": {
     "duration": 22.086752,
     "end_time": "2022-05-09T17:46:54.411972",
     "exception": false,
     "start_time": "2022-05-09T17:46:32.325220",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(40006, 3072) (40006, 1)\n",
      "(40006, 3, 32, 32)\n",
      "[0]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 108x108 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Importing the Labelled Dataset\n",
    "df_train = pd.read_csv(\"../input/cifar10/train_lab_x.csv\")\n",
    "y_train = pd.read_csv(\"../input/cifar10/train_lab_y.csv\")\n",
    "df_train = np.array(df_train)\n",
    "y_train = np.array(y_train)\n",
    "print(df_train.shape, y_train.shape)\n",
    "\n",
    "# Reshaping the dataset\n",
    "df_train = np.reshape(df_train, (-1, 3, 32, 32))\n",
    "print(df_train.shape)\n",
    "\n",
    "# Visualizing a single image\n",
    "ind = 11\n",
    "example = df_train[ind, : , : , : ]\n",
    "example = example.transpose((1, 2, 0))\n",
    "plt.figure(figsize=(1.5, 1.5))\n",
    "plt.imshow(example)\n",
    "print(y_train[ind])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a43cf16f",
   "metadata": {
    "papermill": {
     "duration": 0.016353,
     "end_time": "2022-05-09T17:46:54.445540",
     "exception": false,
     "start_time": "2022-05-09T17:46:54.429187",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 3. Basic Pre-processing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "d4d0db48",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:54.486930Z",
     "iopub.status.busy": "2022-05-09T17:46:54.485915Z",
     "iopub.status.idle": "2022-05-09T17:46:56.897812Z",
     "shell.execute_reply": "2022-05-09T17:46:56.898750Z",
     "shell.execute_reply.started": "2022-05-09T17:45:15.324067Z"
    },
    "papermill": {
     "duration": 2.435209,
     "end_time": "2022-05-09T17:46:56.898954",
     "exception": false,
     "start_time": "2022-05-09T17:46:54.463745",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(40006, 32, 32, 3) (40006, 10)\n"
     ]
    }
   ],
   "source": [
    "# Creating a random permutation\n",
    "perm = np.random.permutation(df_train.shape[0])\n",
    "\n",
    "# Shuffling the training dataset\n",
    "df_train = df_train[perm, : , : , : ]\n",
    "y_train = y_train[perm]\n",
    "\n",
    "# Reshaping, rescaling and one-hot encoding\n",
    "df_train = np.transpose(np.array(df_train), (0, 2, 3, 1))\n",
    "df_train = df_train / 255\n",
    "y_train_oh = tf.one_hot(np.ravel(y_train), depth = 10)\n",
    "\n",
    "print(df_train.shape, y_train_oh.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e0578242",
   "metadata": {
    "papermill": {
     "duration": 0.016326,
     "end_time": "2022-05-09T17:46:56.932651",
     "exception": false,
     "start_time": "2022-05-09T17:46:56.916325",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 4. Creating the Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "962e86ba",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:56.976456Z",
     "iopub.status.busy": "2022-05-09T17:46:56.975670Z",
     "iopub.status.idle": "2022-05-09T17:46:56.977638Z",
     "shell.execute_reply": "2022-05-09T17:46:56.978062Z",
     "shell.execute_reply.started": "2022-05-09T17:45:17.820638Z"
    },
    "papermill": {
     "duration": 0.029082,
     "end_time": "2022-05-09T17:46:56.978188",
     "exception": false,
     "start_time": "2022-05-09T17:46:56.949106",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "def cnn_model(input_shape):\n",
    "    input_img = tf.keras.Input(shape = input_shape)\n",
    "    \n",
    "    Z1 = tfl.Conv2D(16, kernel_size=3, strides=1, padding='same', activation='relu')(input_img)\n",
    "    B1 = tfl.BatchNormalization(axis=-1)(Z1)\n",
    "    Z2 = tfl.Conv2D(16, kernel_size=3, strides=1, padding='same', activation='relu')(B1)\n",
    "    B2 = tfl.BatchNormalization(axis=-1)(Z2)\n",
    "    P1 = tfl.MaxPool2D(pool_size=2, strides=2, padding='valid')(B2)\n",
    "    D1 = tfl.Dropout(0.25)(P1)\n",
    "    # (16, 16, 16)\n",
    "    \n",
    "    Z3 = tfl.Conv2D(32, kernel_size=2, strides=1, padding='valid', activation='relu')(D1)\n",
    "    B3 = tfl.BatchNormalization(axis=-1)(Z3)\n",
    "    Z4 = tfl.Conv2D(32, kernel_size=2, strides=1, padding='valid', activation='relu')(B3)\n",
    "    B4 = tfl.BatchNormalization(axis=-1)(Z4)\n",
    "    P2 = tfl.MaxPool2D(pool_size=2, strides=2, padding='valid')(B4)\n",
    "    D2 = tfl.Dropout(0.25)(P2)\n",
    "    # (7, 7, 32)\n",
    "    \n",
    "    F1 = tfl.Flatten()(D2)\n",
    "    Den1 = tfl.Dense(256, activation='relu')(F1)\n",
    "    Drop1 = tfl.Dropout(0.25)(Den1)\n",
    "    Den2 = tfl.Dense(64, activation='relu')(Drop1)\n",
    "    Drop2 = tfl.Dropout(0.25)(Den2)\n",
    "    outputs = tfl.Dense(10, activation='softmax')(Drop2)\n",
    "    \n",
    "    model = tf.keras.Model(inputs = input_img, outputs = outputs)\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "51a66d05",
   "metadata": {
    "papermill": {
     "duration": 0.016231,
     "end_time": "2022-05-09T17:46:57.011196",
     "exception": false,
     "start_time": "2022-05-09T17:46:56.994965",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "# 5. Compiling, Training, Hyperparameter-Tuning and Evaluating the Model\n",
    "## 5.1. Compiling the Model & Datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "3e30e88f",
   "metadata": {
    "_kg_hide-output": true,
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:57.048718Z",
     "iopub.status.busy": "2022-05-09T17:46:57.048201Z",
     "iopub.status.idle": "2022-05-09T17:46:57.208454Z",
     "shell.execute_reply": "2022-05-09T17:46:57.207757Z",
     "shell.execute_reply.started": "2022-05-09T17:45:17.837521Z"
    },
    "papermill": {
     "duration": 0.180473,
     "end_time": "2022-05-09T17:46:57.208580",
     "exception": false,
     "start_time": "2022-05-09T17:46:57.028107",
     "status": "completed"
    },
    "scrolled": true,
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 32, 32, 3)]       0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 32, 32, 16)        448       \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 32, 32, 16)        64        \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 32, 32, 16)        2320      \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 32, 32, 16)        64        \n",
      "_________________________________________________________________\n",
      "max_pooling2d (MaxPooling2D) (None, 16, 16, 16)        0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 16, 16, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 15, 15, 32)        2080      \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 15, 15, 32)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 14, 14, 32)        4128      \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 14, 14, 32)        128       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 7, 7, 32)          0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 7, 7, 32)          0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 1568)              0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 256)               401664    \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                16448     \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 428,122\n",
      "Trainable params: 427,930\n",
      "Non-trainable params: 192\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "conv_model = cnn_model((32, 32, 3))\n",
    "conv_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics='accuracy')\n",
    "conv_model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "d724e93a",
   "metadata": {
    "_kg_hide-output": true,
    "execution": {
     "iopub.execute_input": "2022-05-09T17:46:57.254335Z",
     "iopub.status.busy": "2022-05-09T17:46:57.253413Z",
     "iopub.status.idle": "2022-05-09T17:47:04.762844Z",
     "shell.execute_reply": "2022-05-09T17:47:04.763531Z",
     "shell.execute_reply.started": "2022-05-09T17:45:17.979338Z"
    },
    "papermill": {
     "duration": 7.536172,
     "end_time": "2022-05-09T17:47:04.763742",
     "exception": false,
     "start_time": "2022-05-09T17:46:57.227570",
     "status": "completed"
    },
    "scrolled": true,
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 3072) (10000, 1)\n",
      "(10000, 3, 32, 32)\n",
      "(10000, 32, 32, 3) (10000, 10)\n"
     ]
    }
   ],
   "source": [
    "# Composing the Train Dataset\n",
    "train_dataset = tf.data.Dataset.from_tensor_slices((df_train, y_train_oh)).batch(32)\n",
    "\n",
    "# Importing the Test Dataset\n",
    "df_test = pd.read_csv(\"../input/cifar10/test_x.csv\")\n",
    "y_test = pd.read_csv(\"../input/cifar10/test_y.csv\")\n",
    "df_test = np.array(df_test)\n",
    "y_test = np.array(y_test)\n",
    "print(df_test.shape, y_test.shape)\n",
    "\n",
    "# Reshaping the dataset\n",
    "df_test = np.reshape(df_test, (-1, 3, 32, 32))\n",
    "print(df_test.shape)\n",
    "\n",
    "# Reshaping, rescaling and one-hot encoding\n",
    "df_test = np.transpose(np.array(df_test), (0, 2, 3, 1))\n",
    "df_test = df_test / 255\n",
    "y_test_oh = tf.one_hot(np.ravel(y_test), depth = 10)\n",
    "print(df_test.shape, y_test_oh.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "79727909",
   "metadata": {
    "papermill": {
     "duration": 0.019662,
     "end_time": "2022-05-09T17:47:04.804509",
     "exception": false,
     "start_time": "2022-05-09T17:47:04.784847",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "## 5.2. Training & Hyperparameter-Tuning for the Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "e9bdeb35",
   "metadata": {
    "_kg_hide-output": true,
    "execution": {
     "iopub.execute_input": "2022-05-09T17:47:04.859270Z",
     "iopub.status.busy": "2022-05-09T17:47:04.852114Z",
     "iopub.status.idle": "2022-05-09T17:47:05.022515Z",
     "shell.execute_reply": "2022-05-09T17:47:05.021619Z",
     "shell.execute_reply.started": "2022-05-09T17:45:25.055813Z"
    },
    "papermill": {
     "duration": 0.198447,
     "end_time": "2022-05-09T17:47:05.022659",
     "exception": false,
     "start_time": "2022-05-09T17:47:04.824212",
     "status": "completed"
    },
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "# If the model has been pre-trained\n",
    "try:\n",
    "    conv_model = cnn_model((32, 32, 3))\n",
    "    conv_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics='accuracy')\n",
    "    conv_model.load_weights(\"../input/dcai-rw/baseline_model.h5\")\n",
    "\n",
    "# If the model hasn't been pre-trained\n",
    "except:\n",
    "    num_epochs = [10, 20, 30, 40, 50]\n",
    "    train_loss, test_loss, train_acc, test_acc = [], [], [], []\n",
    "\n",
    "    for epochs in num_epochs:\n",
    "        # Training the Model\n",
    "        conv_model = cnn_model((32, 32, 3))\n",
    "        conv_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics='accuracy')\n",
    "        conv_model.fit(train_dataset, epochs = epochs)\n",
    "\n",
    "        # Predicting on the Train/Test Datasets\n",
    "        preds_train = conv_model.predict(df_train)\n",
    "        preds_test = conv_model.predict(df_test)\n",
    "\n",
    "        # Finding the Predicted Classes\n",
    "        cls_train = np.argmax(preds_train, axis = 1)\n",
    "        cls_test = np.argmax(preds_test, axis = 1)\n",
    "\n",
    "        # Finding the Train/Test set Loss\n",
    "        train_loss.append(log_loss(y_train_oh, preds_train))\n",
    "        test_loss.append(log_loss(y_test_oh, preds_test))\n",
    "        train_acc.append(accuracy_score(y_train, cls_train))\n",
    "        test_acc.append(accuracy_score(y_test, cls_test))\n",
    "\n",
    "        print(\"For \", epochs, \" Epochs:\")\n",
    "        print(\"Log-loss for Train Dataset = \", train_loss[-1])\n",
    "        print(\"Log-loss for Test Dataset = \", test_loss[-1])\n",
    "        print(\"Accuracy for Train Dataset = \", train_acc[-1])\n",
    "        print(\"Accuracy for Test Dataset = \", test_acc[-1])\n",
    "        print()\n",
    "    \n",
    "    # Training the Model with the best hyper-parameter settings\n",
    "    ind = np.argmax(test_acc)\n",
    "    best_num_epochs = num_epochs[ind]\n",
    "    conv_model = cnn_model((32, 32, 3))\n",
    "    conv_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics='accuracy')\n",
    "    conv_model.fit(train_dataset, epochs = best_num_epochs)\n",
    "\n",
    "    # Saving the model along with it's weights\n",
    "    conv_model.save('baseline_model.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06284a76",
   "metadata": {
    "papermill": {
     "duration": 0.01905,
     "end_time": "2022-05-09T17:47:05.061461",
     "exception": false,
     "start_time": "2022-05-09T17:47:05.042411",
     "status": "completed"
    },
    "tags": []
   },
   "source": [
    "## 5.3. Evaluating/Inferencing on the Train/Test Sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "1aea9ff2",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:47:05.105203Z",
     "iopub.status.busy": "2022-05-09T17:47:05.104361Z",
     "iopub.status.idle": "2022-05-09T17:47:17.438057Z",
     "shell.execute_reply": "2022-05-09T17:47:17.438549Z",
     "shell.execute_reply.started": "2022-05-09T17:45:25.251939Z"
    },
    "papermill": {
     "duration": 12.358028,
     "end_time": "2022-05-09T17:47:17.438725",
     "exception": false,
     "start_time": "2022-05-09T17:47:05.080697",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "# Predicting on the Train/Test Datasets\n",
    "preds_train = conv_model.predict(df_train)\n",
    "preds_test = conv_model.predict(df_test)\n",
    "\n",
    "# Finding the Predicted Classes\n",
    "cls_train = np.argmax(preds_train, axis = 1)\n",
    "cls_test = np.argmax(preds_test, axis = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "8374ccf4",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-05-09T17:47:17.483326Z",
     "iopub.status.busy": "2022-05-09T17:47:17.482726Z",
     "iopub.status.idle": "2022-05-09T17:47:17.585457Z",
     "shell.execute_reply": "2022-05-09T17:47:17.586144Z",
     "shell.execute_reply.started": "2022-05-09T17:45:30.604035Z"
    },
    "papermill": {
     "duration": 0.12757,
     "end_time": "2022-05-09T17:47:17.586354",
     "exception": false,
     "start_time": "2022-05-09T17:47:17.458784",
     "status": "completed"
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Log-loss for Train Dataset =  0.07668840324139689\n",
      "Log-loss for Test Dataset =  0.780156585093371\n",
      "Weighted F1 Score for Train Dataset =  0.9808037008857216\n",
      "Weighted F1 Score for Test Dataset =  0.7805332999350915\n",
      "Accuracy for Train Dataset =  0.9808528720691896\n",
      "Accuracy for Test Dataset =  0.7822\n"
     ]
    }
   ],
   "source": [
    "# Finding the Train/Test set Loss\n",
    "print(\"Log-loss for Train Dataset = \", log_loss(y_train_oh, preds_train))\n",
    "print(\"Log-loss for Test Dataset = \", log_loss(y_test_oh, preds_test))\n",
    "print(\"Weighted F1 Score for Train Dataset = \", f1_score(y_train, cls_train, average = 'weighted'))\n",
    "print(\"Weighted F1 Score for Test Dataset = \", f1_score(y_test, cls_test, average = 'weighted'))\n",
    "print(\"Accuracy for Train Dataset = \", accuracy_score(y_train, cls_train))\n",
    "print(\"Accuracy for Test Dataset = \", accuracy_score(y_test, cls_test))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.12"
  },
  "papermill": {
   "default_parameters": {},
   "duration": 65.329914,
   "end_time": "2022-05-09T17:47:20.850934",
   "environment_variables": {},
   "exception": null,
   "input_path": "__notebook__.ipynb",
   "output_path": "__notebook__.ipynb",
   "parameters": {},
   "start_time": "2022-05-09T17:46:15.521020",
   "version": "2.3.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}