def test_lookback_window_forecast_valueerror_during_predict(self):
# Assert that (for LSTMForecast) ValueError is raised
# in fit method if lookback_window >= number of readings (rows of X)
X = np.random.random(size=(5, 2))
for lookback_window in [5, 6]:
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model,
lookback_window=lookback_window)
with self.assertRaises(ValueError):
model.fit(X)
def test_lookback_window_forecast_valueerror_during_fit(lookback_window):
"""
Assert (for LSTMForecast) ValueError is raised
in fit method if lookback_window >= number of readings (rows of X)
"""
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model,
lookback_window=lookback_window)
with pytest.raises(ValueError):
X = np.random.random(size=(5, 2))
y = X.copy()
model.fit(X, y)
def test_lstmforecast_transform_output(self):
# test for KerasLSTMForecast
# - test dimension of output
# - test that first half of output is testing data
X_train = np.random.random(size=(5, 3))
lookback_window = 3
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model,
lookback_window=lookback_window)
model = model.fit(X_train)
X_test = np.random.random(size=(6, 3))
out = model.transform(X_test)
self.assertEqual(out.shape, (3, 6))
self.assertEqual(out[:, :3].tolist(),
X_test[lookback_window:, :].tolist())
def test_generate_window_lstm_forecast_output(self):
# Check that right output is generated from generate_window (for KerasLSTMForecast)
X = np.random.random(size=(5, 2))
lookback_window = 3
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model,
lookback_window=lookback_window)
gen = model.generate_window(X)
gen_out_1 = next(gen)
gen_out_2 = next(gen)
self.assertEqual(gen_out_1[0].tolist(), X[0:3].reshape(1, 3,
2).tolist())
self.assertEqual(gen_out_2[0].tolist(), X[1:4].reshape(1, 3,
2).tolist())
self.assertEqual(gen_out_1[1].tolist(), X[3].reshape(1, 2).tolist())
self.assertEqual(gen_out_2[1].tolist(), X[4].reshape(1, 2).tolist())
X = np.random.random(size=(4, 2))
lookback_window = 2
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model,
lookback_window=lookback_window)
gen_no_y = model.generate_window(X, output_y=False)
gen_no_y_out_1 = next(gen_no_y)
gen_no_y_out_2 = next(gen_no_y)
self.assertEqual(gen_no_y_out_1.tolist(), X[0:2].reshape(1, 2,
2).tolist())
self.assertEqual(gen_no_y_out_2.tolist(), X[1:3].reshape(1, 2,
2).tolist())
def test_keras_forecast_reshapes_array(self):
# Asserts that KerasLSTMForecast accepts an array of elements, which it will
# reshape into the matrix of single elements it needs
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model)
X = np.random.rand(100)
model.fit(X)
X = np.random.rand(100)
model.predict(X)
def test_keras_forecast_reshapes_array():
"""
Asserts KerasLSTMForecast accepts an array of elements, which it will
reshape into the matrix of single elements it needs
"""
model = KerasLSTMForecast(kind=lstm_autoencoder.lstm_model)
X, y = np.random.rand(100), np.random.rand(100)
model.fit(X, y)
model.predict(X)