params = {'dim': (input_size,),
'batch_size': config["training"]["batch_size"],
'n_classes': config["data"]["n_classes"],
'shuffle': True}
# Datasets
with open(config["data"]["save_ratings"], 'rb') as fp:
labels = pickle.load(fp)
# Generators
training_generator = DataGenerator(train_indices, labels, config["data"]["save_reviews"], **params)
validation_generator = DataGenerator(valid_indices, labels, config["data"]["save_reviews"], **params)
# Define model
model = CharCNN(input_size=input_size,
alphabet_size=alphabet_size,
embedding_size=config["char_cnn"]["embedding_size"],
conv_layers=config["char_cnn"]["conv_layers"],
fully_connected_layers=config["char_cnn"]["fully_connected_layers"],
n_classes=config["data"]["n_classes"],
threshold=config["char_cnn"]["threshold"],
dropout_p=config["char_cnn"]["dropout_p"],
optimizer=config["char_cnn"]["optimizer"],
loss=config["char_cnn"]["loss"])
# Train model
model.train(training_gen=training_generator,
validation_gen=validation_generator,
epochs=config["training"]["epochs"],
batch_size=config["training"]["batch_size"],
log_freq=config["training"]["log_freq"])
dataTest = Data(list(zip(contents_test, labels_test)), alphabet,
input_size)
test_data, test_labels = dataTest.convert_data()
# Initialize the model
model = CharCNN(input_sz=config["data"]["input_size"],
alphabet_sz=config["data"]["alphabet_size"],
emb_sz=config["char_cnn_zhang"]["embedding_size"],
conv_layers=config["char_cnn_zhang"]["conv_layers"],
fc_layers=[],
threshold=config["char_cnn_zhang"]["threshold"],
dropout_p=config["char_cnn_zhang"]["dropout_p"],
optimizer=config["char_cnn_zhang"]["optimizer"],
loss=config["char_cnn_zhang"]["loss"])
# Train
model.train(train_inputs=train_data,
train_labels=train_labels,
val_inputs=val_data,
val_labels=val_labels,
epochs=config["training"]["epochs"],
bs=config["training"]["batch_size"])
# Evaluate
results = model.test(test_data, test_labels, bs=128)
model.test_model(test_data, test_labels, bs=128)
print(results)
# Save the model
model.model.save('model.h5')