training.py

import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten, Conv2D, MaxPool2D
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import matplotlib.pyplot as plt
from datetime import datetime
import pickle
import os

DATA_DIR = './data/'

# load the pre-processed data
features = pickle.load(open(os.path.join(DATA_DIR, 'features.pickle'), 'rb'))
labels = pickle.load(open(os.path.join(DATA_DIR, 'labels.pickle'), 'rb'))

test_features = pickle.load(open(os.path.join(DATA_DIR, 'test_features.pickle'), 'rb'))
test_labels = pickle.load(open(os.path.join(DATA_DIR, 'test_labels.pickle'), 'rb'))

# normalize it
features = features / 255.0
test_features = test_features / 255.0

# Build the model
model = Sequential()

# 3 convolutional layers with 2 max pooling layers
model.add(Conv2D(50, (3, 3), strides=(1, 1), padding='same', activation='relu', input_shape=features.shape[1:]))

model.add(Conv2D(75, (3, 3), strides=(1, 1), padding='same', activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

model.add(Conv2D(125, (3, 3), strides=(1, 1), padding='same', activation='relu'))
model.add(MaxPool2D(pool_size=(2, 2)))
model.add(Dropout(0.25))

# 2 hidden layers
model.add(Flatten())

model.add(Dense(500, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(250, activation='relu'))
model.add(Dropout(0.2))

# The output layer with 6 neurons, one for each emotion
model.add(Dense(6, activation='softmax'))

# compile the model
opt = tf.keras.optimizers.Adam(learning_rate=0.001)
model.compile(loss='sparse_categorical_crossentropy',
				optimizer=opt,
				metrics=['accuracy'])

# train the model, then evaluate it
history = model.fit(features, labels, batch_size=128, validation_split=0.2, epochs=5)
evaluation = model.evaluate(x=test_features, y=test_labels)

# print out the results of the evaluation
print(model.metrics_names)
print(evaluation)

# save the model
now = datetime.now()
model.save(now.strftime('%d-%m-%Y_%H-%M-%S'))

# create a graph showing the accuracy changes during the training phase
print(history.history.keys())
fig = plt.figure(1)
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('model accuracy')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'validation'], loc='upper left')
fig.savefig('model accuracy.png')