early_stopping_callback = EarlyStoppingCallback(monitor='accuracy',
min_delta=0,
patience=patience,
verbose=verbose)
learning_rate_callback = LearningRateCallback(monitor='accuracy',
min_delta=0,
patience=patience_lr,
verbose=verbose)
callbacks = [
metrics_callback, checkpoint_callback, early_stopping_callback,
learning_rate_callback
]
# Generate the architecture
model = model_generate(2)
# Configure the learning process by compiling the network
model.compile(optimizer='adam', loss='categorical_crossentropy')
print(model.summary())
# Train the model for a fixed number of epochs
model.fit(X_train,
Y_train,
batch_size=batch,
verbose=verbose,
callbacks=callbacks,
validation_data=(X_test, Y_test),
epochs=epochs)
Y_valid_age = '../data/Age/Valid/final_labels_data.npy'
Y_valid_age = np.load(Y_valid_age)
if __name__ == '__main__':
# parameters
batch = 256
epochs = 10000
verbose = 1
input_shape = (64, 64, 1)
learning_rate = 0.01
decay = 1e-6
momentum = 0.9
# Generate the architecture
emotion_model, gender_model, age_model = model_generate()
# Stochastic gradient descent optimizer
sgd = optimizers.SGD(lr=learning_rate,
decay=decay,
momentum=momentum,
nesterov=True)
# Configure the learning process by compiling the network
emotion_model.compile(optimizer=sgd, loss='categorical_crossentropy')
gender_model.compile(optimizer='adam', loss='categorical_crossentropy')
age_model.compile(optimizer='adam', loss='mean_absolute_error')
emotion_accuracy = -math.inf
gender_accuracy = -math.inf
age_mae = math.inf