batch_size = 62
####################
# reshape training into [samples, timesteps, features]
train_X, train_y = train[:, 0:n_lag], train[:, n_lag:]
train_X = train_X.reshape(train_X.shape[0], n_lag, features)
test_X, test_y = test[:, 0:n_lag], test[:, n_lag:]
test_X = test_X.reshape(test_X.shape[0], n_lag, features)
# design network
# in stateful model one should but batch_size into input shape and the same value also should be put on batch_size down
model = Sequential()
model.add(
LSTM(512,
batch_input_shape=(batch_size, train_X.shape[1], train_X.shape[2]),
activation='tanh',
stateful=True,
return_sequences=False))
model.add(Dropout(0.2))
model.add(Dense(n_seq))
model.add(Activation('linear'))
#model.add(LeakyReLU())
model.compile(loss='mse', optimizer='adam')
# fit network
model.fit(train_X,
train_y,
epochs=1,
batch_size=batch_size,
validation_data=(test_X, test_y),
verbose=2,
shuffle=False)
y_test = test_sc_df.dropna().drop('X_1', axis=1)
X_train = X_train.as_matrix()
y_train = y_train.as_matrix()
X_test = X_test.as_matrix()
y_test = y_test.as_matrix()
print('Train size: (%d x %d)' % (X_train.shape[0], X_train.shape[1]))
print('Test size: (%d x %d)' % (X_test.shape[0], X_test.shape[1]))
regressor = SVR(kernel='rbf')
regressor.fit(X_train, y_train)
y_pred = regressor.predict(X_test)
r2_test = mean_squared_error(y_test, y_pred)
K.clear_session()
model = Sequential()
model.add(
Dense(50,
input_shape=(X_test.shape[1], ),
activation='relu',
kernel_initializer='lecun_uniform'))
model.add(Dense(50, input_shape=(X_test.shape[1], ), activation='relu'))
model.add(Dense(1))
model.compile(optimizer=Adam(lr=0.001), loss='mean_squared_error')
model.fit(X_train, y_train, batch_size=12, epochs=24, verbose=0)
y_pred = model.predict(X_test)
print('R-Squared: %f' % (mean_squared_error(y_test, y_pred)))
plt.figure(figsize=(16, 8))
plt.plot(sc.inverse_transform(y_test), label='Resampled')
plt.plot(sc.inverse_transform(y_pred), label='Forecast')
plt.legend(loc='best')
plt.show()
plot = False
else:
