Fix/RegressionEnsemble.extreme_lags when models use only covariates lags

CHANGELOG.md

@@ -12,6 +12,7 @@ but cannot always guarantee backwards compatibility. Changes that may **break co
 
 **Fixed**
 - Fixed a bug in `TimeSeries.from_dataframe()` when using a pandas.DataFrame with `df.columns.name != None`. [#1938](https://github.com/unit8co/darts/pull/1938) by [Antoine Madrona](https://github.com/madtoinou).
+- Fixed a bug in `RegressionEnsembleModel.extreme_lags` when the forecasting models have only covariates lags. [#1942](https://github.com/unit8co/darts/pull/1942) by [Antoine Madrona](https://github.com/madtoinou).
 
 
 ## [0.25.0](https://github.com/unit8co/darts/tree/0.25.0) (2023-08-04)

darts/models/forecasting/regression_ensemble_model.py

@@ -130,8 +130,8 @@ def fit(
 
         raise_if(
             train_n_points_too_big,
-            "`regression_train_n_points` parameter too big (must be smaller or "
-            "equal to the number of points in training_series)",
+            "`regression_train_n_points` parameter too big (must be strictly smaller than "
+            "the number of points in training_series)",
             logger,
         )
 
@@ -217,7 +217,11 @@ def extreme_lags(
         Optional[int],
     ]:
         extreme_lags_ = super().extreme_lags
-        return (extreme_lags_[0] - self.train_n_points,) + extreme_lags_[1:]
+        # shift min_target_lag in the past to account for the regression model training set
+        if extreme_lags_[0] is None:
+            return (-self.train_n_points,) + extreme_lags_[1:]
+        else:
+            return (extreme_lags_[0] - self.train_n_points,) + extreme_lags_[1:]
 
     @property
     def output_chunk_length(self) -> int:

darts/tests/models/forecasting/test_regression_ensemble_model.py

@@ -366,18 +366,41 @@ def test_ensemble_models_denoising_multi_input(self):
         self.helper_test_models_accuracy(ensemble, horizon, ts_sum2, ts_cov2, 3)
 
     def test_call_backtest_regression_ensemble_local_models(self):
-        series = tg.sine_timeseries(
-            value_frequency=(1 / 5), value_y_offset=10, length=50
-        )
         regr_train_n = 10
         ensemble = RegressionEnsembleModel(
             [NaiveSeasonal(5), Theta(2, 5)], regression_train_n_points=regr_train_n
         )
-        ensemble.fit(series)
+        ensemble.fit(self.sine_series)
         assert max(m_.min_train_series_length for m_ in ensemble.models) == 10
         # -10 comes from the maximum minimum train series length of all models
         assert ensemble.extreme_lags == (-10 - regr_train_n, 0, None, None, None, None)
-        ensemble.backtest(series)
+        ensemble.backtest(self.sine_series)
+
+    def test_extreme_lags(self):
+        # forecasting models do not use target lags
+        train_n_points = 10
+        model1 = RandomForest(
+            lags_future_covariates=[0],
+        )
+        model2 = RegressionModel(lags_past_covariates=3)
+        model = RegressionEnsembleModel(
+            forecasting_models=[model1, model2],
+            regression_train_n_points=train_n_points,
+        )
+
+        self.assertEqual(model.extreme_lags, (-train_n_points, 0, -3, -1, 0, 0))
+
+        # mix of all the lags
+        model3 = RandomForest(
+            lags_future_covariates=[-2, 5],
+        )
+        model4 = RegressionModel(lags=[-7, -3], lags_past_covariates=3)
+        model = RegressionEnsembleModel(
+            forecasting_models=[model3, model4],
+            regression_train_n_points=train_n_points,
+        )
+
+        self.assertEqual(model.extreme_lags, (-7 - train_n_points, 0, -3, -1, -2, 5))
 
     def test_stochastic_regression_ensemble_model(self):
         quantiles = [0.25, 0.5, 0.75]