Preprocessing and Training a CNN on Fashion MNIST Dataset

# Import necessary libraries import numpy as np import matplotlib.pyplot as plt import seaborn as sns from sklearn.model_selection import train_test_split from sklearn.metrics import confusion_matrix import tensorflow as tf from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, LSTM # a. Get the data (train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.fashion_mnist.load_data() train_yagnam = {'images': train_images, 'labels': train_labels} test_yagnam = {'images': test_images, 'labels': test_labels} # b. Initial Exploration # 1. Display the size of the training and testing dataset print("Size of the training dataset:", len(train_yagnam['images'])) print("Size of the testing dataset:", len(test_yagnam['images'])) # 2. Display the image resolution (dimension) of the input images image_resolution = train_yagnam['images'].shape[1:] print("Image Resolution (Dimension):", image_resolution) # 3. Display the largest pixel value in the dataset using numpy.amax() largest_pixel_value = np.amax(train_yagnam['images']) print("Largest Pixel Value in the Dataset:", largest_pixel_value)

# c. Data Pre-preprocessing # 1. Normalize pixel values to a range between 0-1 train_yagnam['images'] = train_yagnam['images'] / 255.0 test_yagnam['images'] = test_yagnam['images'] / 255.0 # 2. One-hot encode the labels train_yagnam['labels'] = tf.keras.utils.to_categorical(train_yagnam['labels']) test_yagnam['labels'] = tf.keras.utils.to_categorical(test_yagnam['labels']) # 3. Display the shape of train_yagnam['labels'] and test_yagnam['labels'] print("Shape of train_yagnam['labels']:", train_yagnam['labels'].shape) print("Shape of test_yagnam['labels']:", test_yagnam['labels'].shape) # Number of possible labels num_possible_labels = train_yagnam['labels'].shape[1] print("Number of Possible Labels in the Dataset:", num_possible_labels) # d. Visualization # 1. Create a function to display an image with its true label def display_image(image, true_label): plt.imshow(image, cmap='gray') plt.title(f'True Label: {true_label}') plt.axis('off') plt.show() # 2. Plot the first 12 data samples in the training dataset fig, axes = plt.subplots(nrows=4, ncols=3, figsize=(8, 8)) for i in range(12):

ax = axes[i // 3, i % 3] display_image(train_yagnam['images'][i], true_label=np.argmax(train_yagnam['labels'][i])) ax.axis('off') plt.tight_layout() plt.show() # e. Training Data Preparation # 1. Split the training dataset using Sklearn's train_test_split x_train_yagnam, x_val_yagnam, y_train_yagnam, y_val_yagnam = train_test_split( train_yagnam['images'], train_yagnam['labels'], test_size=0.2, random_state=19 ) # f. Build, Train, and Validate CNN Model # Assuming your original image shape is (28, 28) original_image_shape = (28, 28) # Reshape the input data to add a channel dimension x_train_yagnam = x_train_yagnam.reshape(-1, original_image_shape[0], original_image_shape[1], 1) x_val_yagnam = x_val_yagnam.reshape(-1, original_image_shape[0], original_image_shape[1], 1) test_yagnam['images'] = test_yagnam['images'].reshape(-1, original_image_shape[0], original_image_shape[1], 1) # Assuming your labels are already one-hot encoded num_possible_labels = y_train_yagnam.shape[1]

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

# Build the CNN model cnn_model_yagnam = Sequential([ Conv2D(32, (3, 3), activation='relu', input_shape=(original_image_shape[0], original_image_shape[1], 1)), MaxPooling2D(pool_size=(2, 2)), Conv2D(32, (3, 3), activation='relu'), MaxPooling2D(pool_size=(2, 2)), Flatten(), Dense(100, activation='relu'), Dense(num_possible_labels, activation='softmax') ]) # Compile the model cnn_model_yagnam.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy']) # Display a summary of the model cnn_model_yagnam.summary() # Train and validate the model cnn_history_yagnam = cnn_model_yagnam.fit( x_train_yagnam, y_train_yagnam, validation_data=(x_val_yagnam, y_val_yagnam), epochs=8, batch_size=256 ) # Display the Training Vs Validation Accuracy as a line graph plt.plot(cnn_history_yagnam.history['accuracy'], label='Training Accuracy', color='blue')

plt.plot(cnn_history_yagnam.history['val_accuracy'], label='Validation Accuracy', color='green') plt.xlabel('Epoch') plt.ylabel('Accuracy') plt.title('Training Vs Validation Accuracy') plt.legend() plt.show() # Evaluate the cnn model with the test dataset test_loss_yagnam, test_accuracy_yagnam = cnn_model_yagnam.evaluate(test_yagnam['images'], test_yagnam['labels']) print("Test Accuracy:", test_accuracy_yagnam) # 3. Create predictions on the test dataset cnn_predictions_yagnam = cnn_model_yagnam.predict(test_yagnam['images']) # 4. Create a function to plot the probability distribution of predictions def plot_prediction_distribution(true_label, prediction_probs): plt.bar(range(num_possible_labels), prediction_probs, color='blue', label='Predicted') plt.bar(true_label, prediction_probs[true_label], color='green', label='True') plt.xlabel('Label') plt.ylabel('Probability') plt.title('Prediction Probability Distribution') plt.legend() plt.show() # 5. Display the first 4 images from the test dataset with their prediction probability distribution for i in range(4): true_label = np.argmax(test_yagnam['labels'][i]) prediction_probs = cnn_predictions_yagnam[i]

plot_prediction_distribution(true_label, prediction_probs) # 6. Analyze and discuss the prediction probability distribution # 7. Display the confusion matrix cnn_predictions_labels = np.argmax(cnn_predictions_yagnam, axis=1) true_labels = np.argmax(test_yagnam['labels'], axis=1) conf_matrix = confusion_matrix(true_labels, cnn_predictions_labels) # 8. Analyze and discuss the confusion matrix sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues') plt.xlabel('Predicted Label') plt.ylabel('True Label') plt.title('Confusion Matrix') plt.show() # h. Build, Train, Validate, Test, and Analyze RNN Model # Repeat Steps f and g for an RNN model rnn_model_yagnam = Sequential([ # i. Input layer tf.keras.layers.Input(shape=(image_resolution[0], image_resolution[1])), # ii. 1st Layer - LSTM LSTM(128), # iii. Output layer Dense(num_possible_labels, activation='softmax') ]) # 2. Compile the RNN model rnn_model_yagnam.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

# 3. Display a summary of the RNN model rnn_model_yagnam.summary() # 4. Train and validate the RNN model rnn_history_yagnam = rnn_model_yagnam.fit( x_train_yagnam, y_train_yagnam, validation_data=(x_val_yagnam, y_val_yagnam), epochs=8, batch_size=256 ) # 5. Display the Training Vs Validation Accuracy as a line graph for the RNN model plt.plot(rnn_history_yagnam.history['accuracy'], label='Training Accuracy', color='blue') plt.plot(rnn_history_yagnam.history['val_accuracy'], label='Validation Accuracy', color='green') plt.xlabel('Epoch') plt.ylabel('Accuracy') plt.title('Training Vs Validation Accuracy (RNN Model)') plt.legend() plt.show() # 6. Evaluate the RNN model with the test dataset rnn_test_loss, rnn_test_accuracy = rnn_model_yagnam.evaluate(test_yagnam['images'], test_yagnam['labels']) print("RNN Test Accuracy:", rnn_test_accuracy) # 7. Create predictions on the test dataset with the RNN model rnn_predictions_yagnam = rnn_model_yagnam.predict(test_yagnam['images'])

# 8. Display the first 4 images from the test dataset with their prediction probability distribution for the RNN model for i in range(4): true_label_rnn = np.argmax(test_yagnam['labels'][i]) prediction_probs_rnn = rnn_predictions_yagnam[i] plot_prediction_distribution(true_label_rnn, prediction_probs_rnn) # 9. Analyze and discuss the prediction probability distribution for the RNN model # Function to plot the probability distribution of RNN predictions def plot_rnn_prediction_distribution(true_label, prediction_probs_rnn): plt.bar(range(num_possible_labels), prediction_probs_rnn, color='blue', label='Predicted') plt.bar(true_label, prediction_probs_rnn[true_label], color='green', label='True') plt.xlabel('Label') plt.ylabel('Probability') plt.title('RNN Prediction Probability Distribution') plt.legend() plt.show() # Display the first few examples from the RNN predictions for i in range(4): true_label_rnn = np.argmax(test_yagnam['labels'][i]) prediction_probs_rnn = rnn_predictions_yagnam[i] plot_rnn_prediction_distribution(true_label_rnn, prediction_probs_rnn) # 10. Display the confusion matrix for the RNN model rnn_predictions_labels = np.argmax(rnn_predictions_yagnam, axis=1) conf_matrix_rnn = confusion_matrix(true_labels, rnn_predictions_labels)

# 11. Analyze and discuss the confusion matrix for the RNN model sns.heatmap(conf_matrix_rnn, annot=True, fmt='d', cmap='Blues') plt.xlabel('Predicted Label') plt.ylabel('True Label') plt.title('Confusion Matrix (RNN Model)') plt.show()

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Assignment_1_YagnamSukhadia_COMP_263_301239519

Related Documents