removed TSDataset (#397)

etna/pipeline/base.py

@@ -463,7 +463,7 @@ def _forecast_prediction_interval(
         return predictions
 
     @staticmethod
-    def _validate_residuals_for_interval_estimation(backtest_forecasts: TSDataset, residuals: pd.DataFrame):
+    def _validate_residuals_for_interval_estimation(backtest_forecasts: pd.DataFrame, residuals: pd.DataFrame):
         len_backtest, num_segments = residuals.shape
         min_timestamp = backtest_forecasts.index.min()
         max_timestamp = backtest_forecasts.index.max()
@@ -485,11 +485,11 @@ def _add_forecast_borders(
         self, ts: TSDataset, backtest_forecasts: pd.DataFrame, quantiles: Sequence[float], predictions: TSDataset
     ) -> None:
         """Estimate prediction intervals and add to the forecasts."""
-        backtest_forecasts = TSDataset(df=backtest_forecasts, freq=ts.freq)
-        residuals = (
-            backtest_forecasts.loc[:, pd.IndexSlice[:, "target"]]
-            - ts[backtest_forecasts.index.min() : backtest_forecasts.index.max(), :, "target"]
-        )
+        target = ts[backtest_forecasts.index.min() : backtest_forecasts.index.max(), :, "target"]
+        if not backtest_forecasts.index.equals(target.index):
+            raise ValueError("Historical backtest timestamps must match with the original dataset timestamps!")
+
+        residuals = backtest_forecasts.loc[:, pd.IndexSlice[:, "target"]] - target
 
         self._validate_residuals_for_interval_estimation(backtest_forecasts=backtest_forecasts, residuals=residuals)
         sigma = np.nanstd(residuals.values, axis=0)

tests/test_pipeline/test_pipeline.py

@@ -336,6 +336,23 @@ def test_forecast_prediction_interval_not_builtin_with_nans_error(example_tsds,
         _ = pipeline.forecast(prediction_interval=True, quantiles=[0.025, 0.975])
 
 
+@pytest.mark.filterwarnings("ignore: There are NaNs in target on time span from .* to .*")
+@pytest.mark.parametrize("model", (MovingAverageModel(),))
+@pytest.mark.parametrize("stride", (1, 4, 6))
+def test_add_forecast_borders_overlapping_timestamps(example_tsds, model, stride):
+    example_tsds.df.loc[example_tsds.index[-20:-1], pd.IndexSlice["segment_1", "target"]] = None
+
+    pipeline = Pipeline(model=model, transforms=[DateFlagsTransform()], horizon=5)
+    pipeline.fit(example_tsds)
+
+    forecasts = pipeline.get_historical_forecasts(ts=example_tsds, stride=stride)
+
+    with pytest.raises(ValueError, match="Historical backtest timestamps must match"):
+        pipeline._add_forecast_borders(
+            ts=example_tsds, backtest_forecasts=forecasts, quantiles=[0.025, 0.975], predictions=None
+        )
+
+
 def test_forecast_prediction_interval_correct_values(splited_piecewise_constant_ts):
     """Test that the prediction interval for piecewise-constant dataset is correct."""
     train, test = splited_piecewise_constant_ts