Merge branch 'master' into release/1.10.0

CHANGELOG.md

@@ -34,7 +34,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - 
 - 
 - 
-- 
 ### Fixed
 - 
 - 
@@ -61,6 +60,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Change format of holidays for holiday_plot ([#708](https://github.com/tinkoff-ai/etna/pull/708))
 - Make feature selection transforms return columns in inverse_transform([#688](https://github.com/tinkoff-ai/etna/issues/688))
 - Add xticks parameter for plot_periodogram, clip frequencies to be >= 1 ([#706](https://github.com/tinkoff-ai/etna/pull/706))
+- Make TSDataset method to_dataset work with copy of the passed dataframe ([#741](https://github.com/tinkoff-ai/etna/pull/741))
 ### Fixed
 - Fix bug when `ts.plot` does not save figure ([#714](https://github.com/tinkoff-ai/etna/pull/714))
 - Fix bug in plot_clusters ([#675](https://github.com/tinkoff-ai/etna/pull/675))
@@ -70,6 +70,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Fix tiny prediction intervals ([#722](https://github.com/tinkoff-ai/etna/pull/722))
 - Fix deepcopy issue for fitted deepmodel ([#735](https://github.com/tinkoff-ai/etna/pull/735))
 - Fix making backtest if all segments start with NaNs ([#728](https://github.com/tinkoff-ai/etna/pull/728))
+- Fix logging issues with backtest while emp intervals using ([#747](https://github.com/tinkoff-ai/etna/pull/747))
 
 ## [1.9.0] - 2022-05-17
 ### Added

etna/datasets/tsdataset.py

@@ -657,16 +657,17 @@ def to_dataset(df: pd.DataFrame) -> pd.DataFrame:
         2021-01-04           3           8
         2021-01-05           4           9
         """
-        df["timestamp"] = pd.to_datetime(df["timestamp"])
-        df["segment"] = df["segment"].astype(str)
-        feature_columns = df.columns.tolist()
+        df_copy = df.copy(deep=True)
+        df_copy["timestamp"] = pd.to_datetime(df_copy["timestamp"])
+        df_copy["segment"] = df_copy["segment"].astype(str)
+        feature_columns = df_copy.columns.tolist()
         feature_columns.remove("timestamp")
         feature_columns.remove("segment")
-        df = df.pivot(index="timestamp", columns="segment")
-        df = df.reorder_levels([1, 0], axis=1)
-        df.columns.names = ["segment", "feature"]
-        df = df.sort_index(axis=1, level=(0, 1))
-        return df
+        df_copy = df_copy.pivot(index="timestamp", columns="segment")
+        df_copy = df_copy.reorder_levels([1, 0], axis=1)
+        df_copy.columns.names = ["segment", "feature"]
+        df_copy = df_copy.sort_index(axis=1, level=(0, 1))
+        return df_copy
 
     def _find_all_borders(
         self,

etna/pipeline/base.py

@@ -223,7 +223,8 @@ def _forecast_prediction_interval(
         """Add prediction intervals to the forecasts."""
         if self.ts is None:
             raise ValueError("Pipeline is not fitted! Fit the Pipeline before calling forecast method.")
-        _, forecasts, _ = self.backtest(ts=self.ts, metrics=[MAE()], n_folds=n_folds)
+        with tslogger.disable():
+            _, forecasts, _ = self.backtest(ts=self.ts, metrics=[MAE()], n_folds=n_folds)
         forecasts = TSDataset(df=forecasts, freq=self.ts.freq)
         residuals = (
             forecasts.loc[:, pd.IndexSlice[:, "target"]]

tests/test_datasets/test_dataset.py

@@ -814,3 +814,11 @@ def test_inverse_transform_back_included_columns(ts_with_features, columns, retu
     ts_with_features.fit_transform([transform])
     ts_with_features.inverse_transform()
     assert set(original_regressors) == set(ts_with_features.regressors)
+
+
+def test_to_dataset_not_modify_dataframe():
+    timestamp = pd.date_range("2021-01-01", "2021-02-01")
+    df_original = pd.DataFrame({"timestamp": timestamp, "target": 11, "segment": 1})
+    df_copy = df_original.copy(deep=True)
+    df_mod = TSDataset.to_dataset(df_original)
+    pd.testing.assert_frame_equal(df_original, df_copy)

tests/test_loggers/test_file_logger.py

@@ -258,6 +258,38 @@ def test_local_file_logger_with_stacking_ensemble(example_df):
         tslogger.remove(idx)
 
 
+def test_local_file_logger_with_empirical_prediction_interval(example_df):
+    """Test that LocalFileLogger correctly works in with empirical predicition intervals via backtest."""
+    with tempfile.TemporaryDirectory() as dirname:
+        cur_dir = pathlib.Path(dirname)
+        logger = LocalFileLogger(experiments_folder=dirname, gzip=False)
+
+        idx = tslogger.add(logger)
+        example_df = TSDataset.to_dataset(example_df)
+        example_df = TSDataset(example_df, freq="1H")
+        pipe = Pipeline(model=NaiveModel(), transforms=[], horizon=2)
+        n_folds = 5
+
+        _ = pipe.backtest(
+            example_df,
+            metrics=[MAE()],
+            n_jobs=4,
+            n_folds=n_folds,
+            forecast_params={"prediction_interval": True},
+        )
+
+        assert len(list(cur_dir.iterdir())) == 1, "we've run one experiment"
+
+        current_experiment_dir = list(cur_dir.iterdir())[0]
+        assert len(list(current_experiment_dir.iterdir())) == 2, "crossval and crossval_results folders"
+
+        assert (
+            len(list((current_experiment_dir / "crossval").iterdir())) == n_folds
+        ), "crossval should have `n_folds` runs"
+
+        tslogger.remove(idx)
+
+
 def test_s3_file_logger_fail_init_endpoint_url(monkeypatch):
     """Test that S3FileLogger can't be created without setting 'endpoint_url' environment variable."""
     monkeypatch.delenv("endpoint_url", raising=False)