Training a CNN with CIFAR-10 dataset

Author: DevOpsThinh

ml/deep_learning/convolutional_neural_network/cnn_cifar10.py

@@ -0,0 +1,71 @@
+# Learner: Nguyen Truong Thinh
+# Contact me: [email protected] || +84393280504
+#
+# Topic: Deep Learning: Convolutional Neural Networks (CNNs)
+#           A CNN that trains on Keras's CIFAR-10 images dataset
+
+import numpy as np
+
+from keras.datasets import cifar10
+from keras.models import Sequential
+from keras.layers import BatchNormalization, Conv2D, Dense, Dropout, Flatten
+from keras.optimizers import Adam
+from keras.utils import to_categorical
+
+from ml.deep_learning.neural_network_keras.decision_boundaries import decision_boundary_2dimensional as boundary
+
+# Hyperparameters we can adjust
+DROPOUT_PROBABILITY = 0.5
+
+(X_train_raw, Y_train_raw), (X_test_raw, Y_test_raw) = cifar10.load_data()
+
+X_train = X_train_raw / 255
+X_test_all = X_test_raw / 255
+X_validation, X_test = np.split(X_test_all, 2)
+Y_train_encoded = to_categorical(Y_train_raw)
+Y_validation_encoded, Y_test = np.split(to_categorical(Y_test_raw), 2)
+
+# Create a sequential model
+model = Sequential()
+
+# Convolutional layers
+model.add(Conv2D(16, (3, 3), activation='relu'))
+model.add(BatchNormalization())  # To improve accuracy
+model.add(Dropout(DROPOUT_PROBABILITY))  # To reduce over-fitting
+
+model.add(Conv2D(32, (3, 3), activation='relu'))
+model.add(BatchNormalization())
+model.add(Dropout(DROPOUT_PROBABILITY))
+
+# Four-dimensional tensor => bi-dimensional matrix of flat data.
+model.add(Flatten())
+
+# Fully connected layers - Dense
+model.add(Dense(1000, activation='relu'))
+model.add(BatchNormalization())
+model.add(Dropout(DROPOUT_PROBABILITY))
+
+model.add(Dense(512, activation='relu'))
+model.add(BatchNormalization())
+model.add(Dropout(DROPOUT_PROBABILITY))
+
+model.add(Dense(10, activation='softmax'))
+
+#  Compile a model
+model.compile(
+    loss='categorical_crossentropy',
+    optimizer=Adam(),
+    metrics=['accuracy']
+)
+
+# Train a network
+history = model.fit(
+    X_train,
+    Y_train_encoded,
+    validation_data=(X_validation, Y_validation_encoded),
+    epochs=20,
+    batch_size=32
+)
+
+# Draw a decision boundary
+boundary.plot(history)

ml/deep_learning/convolutional_neural_network/random_image_samples_from_cifar10.py

@@ -0,0 +1,24 @@
+# Learner: Nguyen Truong Thinh
+# Contact me: [email protected] || +84393280504
+#
+# Topic: Deep Learning: Convolutional Neural Networks (CNNs)
+#           The Keras's CIFAR-10 images dataset
+#           Prints random sample images this dataset.
+
+from keras.datasets import cifar10
+import matplotlib.pyplot as plt
+import random
+
+# Load CIFAR-10
+(X, Y), (_, _) = cifar10.load_data()
+
+# Print a 4x10 grid of images
+ROWS = 4
+COLUMNS = 10
+for i in range(ROWS * COLUMNS):
+    ax = plt.subplot(ROWS, COLUMNS, i + 1)  # Get the next cell in the grid
+    ax.axis('off')                          # Remove ticks on axes
+    idx = random.randint(0, X.shape[0])     # Select a random image
+    ax.set_title(Y[idx][0], fontsize=15)    # Print the image's label
+    ax.imshow(X[idx])                       # Show the image
+plt.show()

ml/deep_learning/convolutional_neural_network/sample_images_cifar10.png

@@ -9,6 +9,7 @@
 
 from ml.deep_learning.keras_functional_api.inception_v4_network import model
 
+# https://drive.google.com/file/d/1NqMs2js-uOyOLL9MB0nzX0AoHx-hy8ty/view?usp=share_link
 weights_path = 'inceptionv4_dogscats_weights.h5'
 model.load_weights(weights_path)