"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### CNN architectures for image classification and data augmentation technique"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import numpy as np\n",
"import matplotlib.pyplot as plt\n",
"from matplotlib.image import imread # matplotlib funtion that is used to load RGB pictured\n",
"\n",
"def load_data(img_path, label_path):\n",
" \"\"\"function that loads the data \"\"\"\n",
" with open(label_path) as file:\n",
"    class_labels = [label.strip(\"\\n\").split('\\t') for label in file.readlines()]\n",
" \n",
" images = []\n",
" labels = []\n",
" for file, label in class_labels:\n",
"    images.append(imread(os.path.join(img_path, file)))\n",
"    labels.append(int(label))\n",
" return np.array(images), np.array(labels)\n",
"\n",
"train_images, train_labels = load_data('assignment3/train', 'assignment3/train_class_labels.txt')\n",
"test_images, test_labels = load_data('assignment3/test', 'assignment3/test_class_labels.txt')\n",
"\n",
"train_images_augmented, train_labels_augmented =\\\n",
"    load_data('assignment3/augmented/train', 'assignment3/augmented/train_class_labels.txt')\n",
"\n",
"# print shapes \n",
"print(\"training size:\", train_images.shape)\n",
"print(\"training class labels size:\", train_labels.shape)\n",
"\n",
"print(\"\\ntest size:\", test_images.shape)\n",
"print(\"test class labels size:\", test_labels.shape)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"from keras.models import Sequential # what kind of model ? a sequenctial model\n",
"from keras.layers.core import Dense, Dropout # different layers, activation function, and dropout\n",
"from keras.optimizers import Adam # optimization algorithm\n",
"from keras.layers.normalization import BatchNormalization\n",
"from keras.layers.convolutional import Conv2D\n",
"from keras.layers.convolutional import MaxPooling2D\n",
"from keras.layers.core import Flatten\n",
"from keras.losses import binary_crossentropy\n",
"\n",
"def build_model(lr, show_summary=False):\n",
" \"\"\" function that builds a keras model using the provided hyperparameters \"\"\"\n",
" model = Sequential(name=f\"CNN model - Learning_rate: {lr}\")\n",
" model.add(\n",
"    Conv2D(32, kernel_size = (3,3), activation=\"relu\", input_shape = (32,32,3))\n",
" )\n",
" model.add(\n",
"    MaxPooling2D(pool_size=(2,2), )\n",
" )\n",
" model.add(BatchNormalization())\n",
" model.add(Dropout(0.2))\n",
" model.add(\n",
"    Conv2D(64, kernel_size = (3,3), activation=\"relu\")\n",
" )\n",
" model.add(\n",
"    MaxPooling2D(pool_size=(2,2

Fit a convolutional neural network to the CIFAR 10 dataset Note:...