def test_singlets_holtwinters(seasonal, h, datatype):
global airpassengers
airpassengers = np.asarray(airpassengers, dtype=datatype)
train = airpassengers[:-h]
test = airpassengers[-h:]
if seasonal == "multiplicative":
pytest.xfail("Statsmodels nan errors with gcc 9.3 (Issue #3384)")
sm_hw = sm_ES(train,
initialization_method='heuristic',
seasonal=seasonal,
seasonal_periods=12)
sm_hw = sm_hw.fit()
# train = cudf.Series(train)
cu_hw = cuml_ES(train, seasonal=seasonal,
seasonal_periods=12)
cu_hw.fit()
cu_pred = cu_hw.forecast(h)
sm_pred = sm_hw.forecast(h)
cu_r2 = r2_score(cu_pred.to_numpy(), test)
sm_r2 = r2_score(sm_pred, test)
assert (cu_r2 >= sm_r2) or (abs(cu_r2 - sm_r2) < 2e-1)
def test_start_freq_holtwinters(frequency, start_periods):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=np.float64)
cu_hw = cuml_ES(data, ts_num=3, seasonal_periods=frequency,
start_periods=start_periods)
cu_hw.fit()
cu_hw.forecast(5)
def test_get_season(freq):
data = np.sin(np.arange(0, 100)*freq)
cu_hw = cuml_ES(data)
cu_hw.fit()
evens = np.arange(0, 98, 2)
odds = evens+1
seasons = cu_hw.get_season().to_numpy()
base = seasons[0]
assert pytest.approx(seasons[evens], 1e-4) == base
assert pytest.approx(seasons[odds], 1e-4) == -1*base
def test_series_holtwinters(idx, h):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=np.float64)
cu_hw = cuml_ES(data, ts_num=3)
cu_hw.fit()
cu_hw.forecast(h, idx)
cu_hw.score(idx)
cu_hw.get_level(idx)
cu_hw.get_trend(idx)
cu_hw.get_season(idx)
def test_inputs_holtwinters(datatype):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=datatype)
cu_hw = cuml_ES(data, ts_num=3)
cu_hw.fit()
cu_hw.forecast(5)
cu_hw.score()
cu_hw.get_level(0)
cu_hw.get_trend(1)
cu_hw.get_season(2)
def test_multits_holtwinters(seasonal, h, datatype):
global airpassengers, co2
airpassengers = np.asarray(airpassengers, dtype=datatype)
co2 = np.asarray(co2, dtype=datatype)
if seasonal == "multiplicative":
pytest.xfail("Statsmodels nan errors with gcc 9.3 (Issue #3384)")
air_train = airpassengers[:-h]
air_test = airpassengers[-h:]
co2_train = co2[:-h]
co2_test = co2[-h:]
data = np.asarray([air_train, co2_train], dtype=datatype)
cu_hw = cuml_ES(data, seasonal=seasonal,
seasonal_periods=12, ts_num=2)
sm_air_hw = sm_ES(air_train,
initialization_method='heuristic',
seasonal=seasonal,
seasonal_periods=12)
sm_co2_hw = sm_ES(co2_train,
initialization_method='heuristic',
seasonal=seasonal,
seasonal_periods=12)
cu_hw.fit()
sm_air_hw = sm_air_hw.fit()
sm_co2_hw = sm_co2_hw.fit()
cu_air_pred = cu_hw.forecast(h, 0)
cu_co2_pred = cu_hw.forecast(h, 1)
sm_air_pred = sm_air_hw.forecast(h)
sm_co2_pred = sm_co2_hw.forecast(h)
cu_air_r2 = r2_score(cu_air_pred.to_numpy(), air_test)
cu_co2_r2 = r2_score(cu_co2_pred.to_numpy(), co2_test)
sm_air_r2 = r2_score(sm_air_pred, air_test)
sm_co2_r2 = r2_score(sm_co2_pred, co2_test)
assert (cu_air_r2 >= sm_air_r2) or (abs(cu_air_r2 - sm_air_r2) < 4)
assert (cu_co2_r2 >= sm_co2_r2) or (abs(cu_co2_r2 - sm_co2_r2) < 4)
full_cu_pred = cu_hw.forecast(h)
air_cu_r2 = r2_score(full_cu_pred[0].to_numpy(), air_test)
co2_cu_r2 = r2_score(full_cu_pred[1].to_numpy(), co2_test)
assert (air_cu_r2 >= sm_air_r2) or (abs(air_cu_r2 - sm_air_r2) < 4)
assert (co2_cu_r2 >= sm_co2_r2) or (abs(co2_cu_r2 - sm_co2_r2) < 4)
def test_multits_holtwinters(seasonal, h, datatype, input_type):
global airpassengers, co2
airpassengers = np.asarray(airpassengers, dtype=datatype)
co2 = np.asarray(co2, dtype=datatype)
air_train = airpassengers[:-h]
air_test = airpassengers[-h:]
co2_train = co2[:-h]
co2_test = co2[-h:]
data = np.asarray([air_train, co2_train], dtype=datatype)
if input_type == 'cudf':
data = cudf.DataFrame({i: data[i] for i in range(data.shape[0])})
cu_hw = cuml_ES(data, seasonal=seasonal,
seasonal_periods=12, ts_num=2)
sm_air_hw = sm_ES(air_train,
seasonal=seasonal,
seasonal_periods=12)
sm_co2_hw = sm_ES(co2_train,
seasonal=seasonal,
seasonal_periods=12)
cu_hw.fit()
sm_air_hw = sm_air_hw.fit()
sm_co2_hw = sm_co2_hw.fit()
cu_air_pred = cu_hw.forecast(h, 0)
cu_co2_pred = cu_hw.forecast(h, 1)
sm_air_pred = sm_air_hw.forecast(h)
sm_co2_pred = sm_co2_hw.forecast(h)
cu_air_r2 = r2_score(cu_air_pred, air_test)
cu_co2_r2 = r2_score(cu_co2_pred, co2_test)
sm_air_r2 = r2_score(sm_air_pred, air_test)
sm_co2_r2 = r2_score(sm_co2_pred, co2_test)
assert (cu_air_r2 >= sm_air_r2) or (abs(cu_air_r2 - sm_air_r2) < 2e-1)
assert (cu_co2_r2 >= sm_co2_r2) or (abs(cu_co2_r2 - sm_co2_r2) < 2e-1)
full_cu_pred = cu_hw.forecast(h)
air_cu_r2 = r2_score(full_cu_pred[0], air_test)
co2_cu_r2 = r2_score(full_cu_pred[1], co2_test)
assert (air_cu_r2 >= sm_air_r2) or (abs(air_cu_r2 - sm_air_r2) < 2e-1)
assert (co2_cu_r2 >= sm_co2_r2) or (abs(co2_cu_r2 - sm_co2_r2) < 2e-1)
def test_inputs_holtwinters(datatype, input_type):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=datatype)
if input_type == 'cudf':
data = cudf.DataFrame({i: data[i] for i in range(data.shape[0])})
elif input_type == 'cupy':
try:
import cupy as cp
data = cp.asarray(data)
except ImportError:
pytest.skip("CuPy import error -- skipping test.")
cu_hw = cuml_ES(data, ts_num=3)
cu_hw.fit()
cu_hw.forecast(5)
cu_hw.score()
cu_hw.get_level(0)
cu_hw.get_trend(1)
cu_hw.get_season(2)
def test_singlets_holtwinters(seasonal, h, datatype):
global airpassengers
airpassengers = np.asarray(airpassengers, dtype=datatype)
train = airpassengers[:-h]
test = airpassengers[-h:]
sm_hw = sm_ES(train, seasonal=seasonal, seasonal_periods=12)
sm_hw = sm_hw.fit()
# train = cudf.Series(train)
cu_hw = cuml_ES(train, seasonal=seasonal, seasonal_periods=12)
cu_hw.fit()
cu_pred = cu_hw.forecast(h)
sm_pred = sm_hw.forecast(h)
cu_r2 = r2_score(cu_pred, test)
sm_r2 = r2_score(sm_pred, test)
assert (cu_r2 >= sm_r2) or (abs(cu_r2 - sm_r2) < 2e-1)
def test_multits_holtwinters(seasonal, h, datatype):
global airpassengers, co2
airpassengers = np.asarray(airpassengers, dtype=datatype)
co2 = np.asarray(co2, dtype=datatype)
air_train = airpassengers[:-h]
air_test = airpassengers[-h:]
co2_train = co2[:-h]
co2_test = co2[-h:]
data = np.asarray([air_train, co2_train], dtype=datatype)
cu_hw = cuml_ES(data, seasonal=seasonal,
seasonal_periods=12, ts_num=2)
sm_air_hw = sm_ES(air_train,
seasonal=seasonal,
seasonal_periods=12)
sm_co2_hw = sm_ES(co2_train,
seasonal=seasonal,
seasonal_periods=12)
cu_hw.fit()
sm_air_hw = sm_air_hw.fit()
sm_co2_hw = sm_co2_hw.fit()
cu_air_pred = cu_hw.forecast(h, 0)
cu_co2_pred = cu_hw.forecast(h, 1)
sm_air_pred = sm_air_hw.forecast(h)
sm_co2_pred = sm_co2_hw.forecast(h)
cu_air_r2 = r2_score(cu_air_pred.to_array(), air_test)
cu_co2_r2 = r2_score(cu_co2_pred.to_array(), co2_test)
sm_air_r2 = r2_score(sm_air_pred, air_test)
sm_co2_r2 = r2_score(sm_co2_pred, co2_test)
assert (cu_air_r2 >= sm_air_r2) or (abs(cu_air_r2 - sm_air_r2) < 4)
assert (cu_co2_r2 >= sm_co2_r2) or (abs(cu_co2_r2 - sm_co2_r2) < 4)
full_cu_pred = cu_hw.forecast(h)
air_cu_r2 = r2_score(full_cu_pred[0], air_test)
co2_cu_r2 = r2_score(full_cu_pred[1], co2_test)
assert (air_cu_r2 >= sm_air_r2) or (abs(air_cu_r2 - sm_air_r2) < 4)
assert (co2_cu_r2 >= sm_co2_r2) or (abs(co2_cu_r2 - sm_co2_r2) < 4)
def test_get_trend(slope):
data = np.arange(0, 100*slope, slope, dtype=np.float64)
cu_hw = cuml_ES(data)
cu_hw.fit()
assert pytest.approx(cu_hw.get_trend().to_numpy(), 1e-4) == slope
def test_get_level(level):
data = np.array([level]*100, dtype=np.float64)
cu_hw = cuml_ES(data)
cu_hw.fit()
assert pytest.approx(cu_hw.get_level().to_numpy(), 1e-4) == level
def test_eps_holtwinters(eps):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=np.float64)
cu_hw = cuml_ES(data, eps=eps, ts_num=3)
cu_hw.fit()
cu_hw.forecast(5)
def test_seasonal_holtwinters(seasonal):
global airpassengers, co2, nybirths
data = np.asarray([airpassengers, co2, nybirths], dtype=np.float64)
cu_hw = cuml_ES(data, seasonal=seasonal, ts_num=3)
cu_hw.fit()
cu_hw.forecast(5)