model.py

import keras


def add_lstm_layer(batch_size, stateful, units, X_shape):
    lstm = keras.layers.LSTM(units, batch_input_shape=(batch_size, X_shape[1], X_shape[2]),
                                 stateful=stateful)

    return lstm


def compile_model(model, optimizer, loss='mean_squared_error'):
    model.compile(loss=loss, optimizer=optimizer)

    return model


def evaluate_model(data, batch_size, model):
    evaluate_model = model.evaluate(data['X_test'], data['y_test'], batch_size=batch_size, verbose=2)

    return evaluate_model


def fit_model(data, batch_size, epochs, model, shuffle):
    fit_model = model.fit(data['X_train'], data['y_train'], batch_size=batch_size, epochs=epochs, shuffle=shuffle,
                        validation_data=(data['X_test'], data['y_test']), verbose=2)

    return fit_model


def make_model(activation, batch_size, data, layers, stateful, units):
    model = keras.models.Sequential()
    model.add(add_lstm_layer(batch_size, stateful, units, data['X_train'].shape))
    # TODO: Test whether return_sequences=True or False makes a difference.

    for layer in layers:
        model.add(layer)

    model.add(activation)

    return model