"""
rmse = np.sqrt(K.mean(K.square(y_pred - y_true))) * 48
return rmse
# Define the name of the weights file that will be trained
weights_file_name = "test002.h5"
feature = ('left_eye_center_x', 'left_eye_center_y', 'right_eye_center_x',
'right_eye_center_y', 'nose_tip_x', 'nose_tip_y',
'mouth_center_bottom_lip_x', 'mouth_center_bottom_lip_y')
flip_indices = [(0, 2), (1, 3)]
# Load dataset using my previous LoadData class
load = LoadData()
X_train, X_val, Y_train, Y_val = load.loadNSplit(feature=feature)
# Define the output number
output_units = Y_train.shape[1]
# Define lenet5 like model
lenet5 = Sequential([
Convolution2D(128, 3, 3, border_mode='valid', input_shape=(1, 96, 96)),
Activation('relu'),
MaxPooling2D(pool_size=(2, 2)),
Dropout(0.1),
Convolution2D(256, 2, 2),
Activation('relu'),
MaxPooling2D(pool_size=(2, 2)),
Dropout(0.25),
Convolution2D(512, 2, 2),
