def test_predict_ts_with_not_X_index(ts_data):
X, y = ts_data
X = X.reset_index(drop=True)
assert not isinstance(X.index, pd.DatetimeIndex)
m_clf = ARIMARegressor()
clf_ = m_clf.fit(X=X, y=y)
with pytest.raises(
ValueError,
match=
"If not it will look for the datetime column in the index of X."):
clf_.predict(X)
def test_fit_predict_ts_with_X_and_y_index(mock_get_dates, mock_format_dates,
ts_data_seasonal):
X, y = ts_data_seasonal
assert isinstance(X.index, pd.DatetimeIndex)
assert isinstance(y.index, pd.DatetimeIndex)
mock_get_dates.return_value = (X.index, X)
mock_format_dates.return_value = (X, y, None)
fh_ = forecasting.ForecastingHorizon(y.index, is_relative=False)
a_clf = sktime_arima.AutoARIMA()
clf = a_clf.fit(X=X, y=y)
y_pred_sk = clf.predict(fh=fh_, X=X)
m_clf = ARIMARegressor(d=None)
m_clf.fit(X=X, y=y)
mock_format_dates.return_value = (X, y, fh_)
y_pred = m_clf.predict(X=X)
assert (y_pred_sk == y_pred.to_series()).all()
def test_fit_predict_ts_with_X_and_y_index(ts_data):
X, y = ts_data
assert isinstance(X.index, pd.DatetimeIndex)
assert isinstance(y.index, pd.DatetimeIndex)
a_clf = arima.ARIMA(endog=y, exog=X, order=(1, 0, 0), trend='n', dates=X.index)
clf = a_clf.fit()
y_pred_a = clf.predict()
m_clf = ARIMARegressor(p=1, d=0, q=0)
clf_ = m_clf.fit(X=X, y=y)
y_pred = clf_.predict(X=X)
assert (y_pred == y_pred_a).all()
def test_fit_predict_date_col_named(ts_data):
X, y = ts_data
a_clf = arima.ARIMA(endog=y, exog=X, order=(1, 0, 0), trend='n', dates=X.index)
clf = a_clf.fit()
y_pred_a = clf.predict(params=(1, 0, 0))
X = X.reset_index()
assert not isinstance(X.index, pd.DatetimeIndex)
m_clf = ARIMARegressor(p=1, d=0, q=0, date_column='index')
clf_ = m_clf.fit(X=X, y=y)
y_pred = clf_.predict(X=X, y=y)
assert (y_pred == y_pred_a).all()
def test_fit_predict_date_index_named_out_of_sample(X_none, ts_data_seasonal):
X, y = ts_data_seasonal
fh_ = forecasting.ForecastingHorizon(y[250:].index, is_relative=False)
a_clf = sktime_arima.AutoARIMA()
if X_none:
clf = a_clf.fit(y=y[:250])
y_pred_sk = clf.predict(fh=fh_)
else:
clf = a_clf.fit(X=X[:250], y=y[:250])
y_pred_sk = clf.predict(fh=fh_, X=X[250:])
X = X.reset_index()
assert not isinstance(X.index, pd.DatetimeIndex)
m_clf = ARIMARegressor(date_index='index', d=None)
if X_none:
m_clf.fit(X=None, y=y[:250])
y_pred = m_clf.predict(X=None, y=y[250:])
else:
m_clf.fit(X=X[:250], y=y[:250])
y_pred = m_clf.predict(X=X[250:], y=y[250:])
assert (y_pred_sk.to_period('D') == y_pred.to_series()).all()
def test_fit_predict_ts_with_y_not_X_index_with_iterable_p_and_q(ts_data):
X, y = ts_data
a_clf = arima.ARIMA(endog=y, exog=X, order=([1, 2], 0, [1, 2]), trend='n', dates=X.index)
clf = a_clf.fit()
y_pred_a = clf.predict(params=(1, 0, 0))
X_no_ind = X.reset_index(drop=True)
assert isinstance(y.index, pd.DatetimeIndex)
assert not isinstance(X_no_ind.index, pd.DatetimeIndex)
m_clf = ARIMARegressor(p=[1, 2], d=0, q=[1, 2])
clf_ = m_clf.fit(X=X_no_ind, y=y)
y_pred = clf_.predict(X=X, y=y)
assert (y_pred == y_pred_a).all()
def test_feature_importance(ts_data):
X, y = ts_data
clf = ARIMARegressor()
clf.fit(X, y)
clf.feature_importance == np.zeros(1)
def test_fit_ts_without_y(ts_data):
X, y = ts_data
clf = ARIMARegressor(p=1, d=0, q=0)
with pytest.raises(ValueError, match="ARIMA Regressor requires y as input."):
clf.fit(X=X)
def test_model_instance(ts_data):
X, y = ts_data
clf = ARIMARegressor()
fitted = clf.fit(X, y)
assert isinstance(fitted, ARIMARegressor)