Make DifferencingTransform to fail on new segments with understandable error

CHANGELOG.md

@@ -32,6 +32,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Fix `SegmentEncoderTransform` to work with subset of segments and raise error on new segments ([#1103](https://github.com/tinkoff-ai/etna/pull/1103))
 - Fix `SklearnTransform` in per-segment mode to work on subset of segments and raise error on new segments ([#1107](https://github.com/tinkoff-ai/etna/pull/1107))
 - Fix `OutliersTransform` and its children to raise error on new segments ([#1139](https://github.com/tinkoff-ai/etna/pull/1139))
+- Fix `DifferencingTransform` to raise error on new segments during `transform` and `inverse_transform` in inplace mode ([#1141](https://github.com/tinkoff-ai/etna/pull/1141))
 ## [1.14.0] - 2022-12-16
 ### Added
 - Add python 3.10 support ([#1005](https://github.com/tinkoff-ai/etna/pull/1005))

etna/transforms/math/differencing.py

@@ -3,11 +3,13 @@
 from typing import Optional
 from typing import Set
 from typing import Union
+from typing import cast
 
 import numpy as np
 import pandas as pd
 
 from etna.transforms.base import Transform
+from etna.transforms.utils import check_new_segments
 from etna.transforms.utils import match_target_quantiles
 
 
@@ -63,6 +65,7 @@ def __init__(
         self.inplace = inplace
         self.out_column = out_column
 
+        self._fit_segments: Optional[List[str]] = None
         self._train_timestamp: Optional[pd.DatetimeIndex] = None
         self._train_init_dict: Optional[Dict[str, pd.Series]] = None
         self._test_init_df: Optional[pd.DataFrame] = None
@@ -101,6 +104,7 @@ def fit(self, df: pd.DataFrame) -> "_SingleDifferencingTransform":
         self._test_init_df = fit_df.iloc[-self.period :, :]
         # make multiindex levels consistent
         self._test_init_df.columns = self._test_init_df.columns.remove_unused_levels()
+        self._fit_segments = df.columns.get_level_values("segment").unique().tolist()
         return self
 
     def transform(self, df: pd.DataFrame) -> pd.DataFrame:
@@ -113,12 +117,9 @@ def transform(self, df: pd.DataFrame) -> pd.DataFrame:
 
         Returns
         -------
-        result: pd.Dataframe
+        result:
             transformed dataframe
         """
-        if self._train_init_dict is None or self._test_init_df is None or self._train_timestamp is None:
-            raise AttributeError("Transform is not fitted")
-
         segments = sorted(set(df.columns.get_level_values("segment")))
         transformed = df.loc[:, pd.IndexSlice[segments, self.in_column]].copy()
         for current_segment in segments:
@@ -207,11 +208,11 @@ def inverse_transform(self, df: pd.DataFrame) -> pd.DataFrame:
 
         Returns
         -------
-        result: pd.DataFrame
+        result:
             transformed DataFrame.
         """
-        if self._train_init_dict is None or self._test_init_df is None or self._train_timestamp is None:
-            raise AttributeError("Transform is not fitted")
+        # we assume this to be fitted
+        self._train_timestamp = cast(pd.DatetimeIndex, self._train_timestamp)
 
         if not self.inplace:
             return df
@@ -312,6 +313,7 @@ def __init__(
             self._differencing_transforms.append(
                 _SingleDifferencingTransform(in_column=result_out_column, period=self.period, inplace=True)
             )
+        self._fit_segments: Optional[List[str]] = None
 
     def _get_column_name(self) -> str:
         if self.inplace:
@@ -337,8 +339,13 @@ def fit(self, df: pd.DataFrame) -> "DifferencingTransform":
         result_df = df.copy()
         for transform in self._differencing_transforms:
             result_df = transform.fit_transform(result_df)
+        self._fit_segments = df.columns.get_level_values("segment").unique().tolist()
         return self
 
+    def _check_is_fitted(self):
+        if self._fit_segments is None:
+            raise ValueError("Transform is not fitted!")
+
     def transform(self, df: pd.DataFrame) -> pd.DataFrame:
         """Make a differencing transformation.
 
@@ -349,9 +356,21 @@ def transform(self, df: pd.DataFrame) -> pd.DataFrame:
 
         Returns
         -------
-        result: pd.Dataframe
+        result:
             transformed dataframe
+
+        Raises
+        ------
+        ValueError:
+            if transform isn't fitted
+        NotImplementedError:
+            if there are segments that weren't present during training
         """
+        self._check_is_fitted()
+        segments = df.columns.get_level_values("segment").unique().tolist()
+        if self.inplace:
+            check_new_segments(transform_segments=segments, fit_segments=self._fit_segments)
+
         result_df = df.copy()
         for transform in self._differencing_transforms:
             result_df = transform.transform(result_df)
@@ -367,12 +386,23 @@ def inverse_transform(self, df: pd.DataFrame) -> pd.DataFrame:
 
         Returns
         -------
-        result: pd.DataFrame
+        result:
             transformed DataFrame.
+
+        Raises
+        ------
+        ValueError:
+            if transform isn't fitted
+        NotImplementedError:
+            if there are segments that weren't present during training
         """
+        self._check_is_fitted()
         if not self.inplace:
             return df
 
+        segments = df.columns.get_level_values("segment").unique().tolist()
+        check_new_segments(transform_segments=segments, fit_segments=self._fit_segments)
+
         result_df = df.copy()
         for transform in self._differencing_transforms[::-1]:
             result_df = transform.inverse_transform(result_df)

etna/transforms/math/sklearn.py

@@ -1,4 +1,3 @@
-import reprlib
 import warnings
 from copy import deepcopy
 from typing import Dict
@@ -14,6 +13,7 @@
 from etna.core import StringEnumWithRepr
 from etna.datasets import set_columns_wide
 from etna.transforms.base import Transform
+from etna.transforms.utils import check_new_segments
 from etna.transforms.utils import match_target_quantiles
 
 
@@ -231,12 +231,8 @@ def _postprocess_macro(self, df: pd.DataFrame, transformed: np.ndarray) -> np.nd
 
     def _preprocess_per_segment(self, df: pd.DataFrame) -> np.ndarray:
         self._fit_segments = cast(List[str], self._fit_segments)
-        transform_segments = df.columns.get_level_values("segment").unique()
-        new_segments = set(transform_segments) - set(self._fit_segments)
-        if len(new_segments) > 0:
-            raise NotImplementedError(
-                f"This transform can't process segments that weren't present on train data: {reprlib.repr(new_segments)}"
-            )
+        transform_segments = df.columns.get_level_values("segment").unique().tolist()
+        check_new_segments(transform_segments=transform_segments, fit_segments=self._fit_segments)
 
         df = df.loc[:, pd.IndexSlice[:, self.in_column]]
         to_add_segments = set(self._fit_segments) - set(transform_segments)

etna/transforms/outliers/base.py

@@ -1,16 +1,15 @@
-import reprlib
 from abc import ABC
 from abc import abstractmethod
 from typing import Dict
 from typing import List
 from typing import Optional
-from typing import cast
 
 import numpy as np
 import pandas as pd
 
 from etna.datasets import TSDataset
 from etna.transforms.base import Transform
+from etna.transforms.utils import check_new_segments
 
 
 class OutliersTransform(Transform, ABC):
@@ -68,14 +67,6 @@ def fit(self, df: pd.DataFrame) -> "OutliersTransform":
 
         return self
 
-    def _validate_segments(self, segments: List[str]):
-        self._fit_segments = cast(List[str], self._fit_segments)
-        new_segments = set(segments) - set(self._fit_segments)
-        if len(new_segments) > 0:
-            raise NotImplementedError(
-                f"This transform can't process segments that weren't present on train data: {reprlib.repr(new_segments)}"
-            )
-
     def transform(self, df: pd.DataFrame) -> pd.DataFrame:
         """
         Replace found outliers with NaNs.
@@ -101,7 +92,7 @@ def transform(self, df: pd.DataFrame) -> pd.DataFrame:
             raise ValueError("Transform is not fitted! Fit the Transform before calling transform method.")
         result_df = df.copy()
         segments = df.columns.get_level_values("segment").unique().tolist()
-        self._validate_segments(segments)
+        check_new_segments(transform_segments=segments, fit_segments=self._fit_segments)
         for segment in segments:
             result_df.loc[self.outliers_timestamps[segment], pd.IndexSlice[segment, self.in_column]] = np.NaN
         return result_df
@@ -131,7 +122,7 @@ def inverse_transform(self, df: pd.DataFrame) -> pd.DataFrame:
             raise ValueError("Transform is not fitted! Fit the Transform before calling inverse_transform method.")
         result = df.copy()
         segments = df.columns.get_level_values("segment").unique().tolist()
-        self._validate_segments(segments)
+        check_new_segments(transform_segments=segments, fit_segments=self._fit_segments)
         for segment in segments:
             segment_ts = result[segment, self.in_column]
             segment_ts[segment_ts.index.isin(self.outliers_timestamps[segment])] = self.original_values[segment]

etna/transforms/utils.py

@@ -1,8 +1,23 @@
 import re
+import reprlib
+from typing import List
+from typing import Optional
 from typing import Set
 
 
 def match_target_quantiles(features: Set[str]) -> Set[str]:
     """Find quantiles in dataframe columns."""
     pattern = re.compile("target_\d+\.\d+$")
     return {i for i in list(features) if pattern.match(i) is not None}
+
+
+def check_new_segments(transform_segments: List[str], fit_segments: Optional[List[str]]):
+    """Check if there are any new segments that weren't present during training."""
+    if fit_segments is None:
+        raise ValueError("Transform is not fitted!")
+
+    new_segments = set(transform_segments) - set(fit_segments)
+    if len(new_segments) > 0:
+        raise NotImplementedError(
+            f"This transform can't process segments that weren't present on train data: {reprlib.repr(new_segments)}"
+        )

pyproject.toml

@@ -188,7 +188,6 @@ line_length = 120
 minversion = "6.0"
 doctest_optionflags = "NORMALIZE_WHITESPACE IGNORE_EXCEPTION_DETAIL NUMBER"
 filterwarnings = [
-    "error",
     "ignore: Torchmetrics v0.9 introduced a new argument class property called `full_state_update` that",
     "ignore: TSDataset freq can't be inferred",
     "ignore: test_size, test_start and test_end cannot be",

tests/test_transforms/test_inference/test_inverse_transform.py

@@ -645,6 +645,7 @@ def test_inverse_transform_train_new_segments(self, transform, dataset_name, exp
             (MeanSegmentEncoderTransform(), "regular_ts"),
             (SegmentEncoderTransform(), "regular_ts"),
             # math
+            (DifferencingTransform(in_column="target", inplace=True), "regular_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=False), "positive_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=True), "positive_ts"),
             (MaxAbsScalerTransform(in_column="target", mode="per-segment", inplace=False), "regular_ts"),
@@ -701,23 +702,6 @@ def test_inverse_transform_train_new_segments_failed_not_implemented(self, trans
             ts, transform, train_segments=["segment_1", "segment_2"], expected_changes={}
         )
 
-    @to_be_fixed(raises=Exception)
-    @pytest.mark.parametrize(
-        "transform, dataset_name, expected_changes",
-        [
-            # math
-            # TODO: error should be understandable, not like now
-            (DifferencingTransform(in_column="target", inplace=True), "regular_ts", {"change": {"target"}}),
-        ],
-    )
-    def test_inverse_transform_train_new_segments_failed_error(
-        self, transform, dataset_name, expected_changes, request
-    ):
-        ts = request.getfixturevalue(dataset_name)
-        self._test_inverse_transform_train_new_segments(
-            ts, transform, train_segments=["segment_1", "segment_2"], expected_changes=expected_changes
-        )
-
 
 class TestInverseTransformFutureNewSegments:
     """Test inverse transform on future part of new segments.
@@ -990,6 +974,8 @@ def test_inverse_transform_future_new_segments(self, transform, dataset_name, ex
             (MeanSegmentEncoderTransform(), "regular_ts"),
             (SegmentEncoderTransform(), "regular_ts"),
             # math
+            (DifferencingTransform(in_column="target", inplace=True), "regular_ts"),
+            (DifferencingTransform(in_column="positive", inplace=True), "ts_with_exog"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=False), "positive_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=True), "positive_ts"),
             (BoxCoxTransform(in_column="positive", mode="per-segment", inplace=True), "ts_with_exog"),
@@ -1080,10 +1066,6 @@ def test_inverse_transform_future_new_segments_failed_not_implemented(self, tran
                 "ts_with_exog",
                 {"create": {"year", "month", "weekday"}},
             ),
-            # math
-            # TODO: error should be understandable, not like now
-            (DifferencingTransform(in_column="target", inplace=True), "regular_ts", {}),
-            (DifferencingTransform(in_column="positive", inplace=True), "ts_with_exog", {"change": {"positive"}}),
         ],
     )
     def test_inverse_transform_future_new_segments_failed_error(

tests/test_transforms/test_inference/test_transform.py

@@ -480,7 +480,6 @@ def _test_transform_train_new_segments(self, ts, transform, train_segments, expe
                 "regular_ts",
                 {"create": {"res"}},
             ),
-            (DifferencingTransform(in_column="target", inplace=True), "regular_ts", {"change": {"target"}}),
             (MADTransform(in_column="target", window=7, out_column="res"), "regular_ts", {"create": {"res"}}),
             (MaxTransform(in_column="target", window=7, out_column="res"), "regular_ts", {"create": {"res"}}),
             (MeanTransform(in_column="target", window=7, out_column="res"), "regular_ts", {"create": {"res"}}),
@@ -601,6 +600,7 @@ def test_transform_train_new_segments(self, transform, dataset_name, expected_ch
             (MeanSegmentEncoderTransform(), "regular_ts"),
             (SegmentEncoderTransform(), "regular_ts"),
             # math
+            (DifferencingTransform(in_column="target", inplace=True), "regular_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=False), "positive_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=True), "positive_ts"),
             (MaxAbsScalerTransform(in_column="target", mode="per-segment", inplace=False), "regular_ts"),
@@ -772,8 +772,6 @@ def _test_transform_future_new_segments(self, ts, transform, train_segments, exp
                 "regular_ts",
                 {"create": {"res"}},
             ),
-            (DifferencingTransform(in_column="target", inplace=True), "regular_ts", {}),
-            (DifferencingTransform(in_column="positive", inplace=True), "ts_with_exog", {"change": {"positive"}}),
             (MADTransform(in_column="target", window=14, out_column="res"), "regular_ts", {"create": {"res"}}),
             (MaxTransform(in_column="target", window=14, out_column="res"), "regular_ts", {"create": {"res"}}),
             (MeanTransform(in_column="target", window=14, out_column="res"), "regular_ts", {"create": {"res"}}),
@@ -924,6 +922,8 @@ def test_transform_future_new_segments(self, transform, dataset_name, expected_c
             (MeanSegmentEncoderTransform(), "regular_ts"),
             (SegmentEncoderTransform(), "regular_ts"),
             # math
+            (DifferencingTransform(in_column="target", inplace=True), "regular_ts"),
+            (DifferencingTransform(in_column="positive", inplace=True), "ts_with_exog"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=False), "positive_ts"),
             (BoxCoxTransform(in_column="target", mode="per-segment", inplace=True), "positive_ts"),
             (BoxCoxTransform(in_column="positive", mode="per-segment", inplace=True), "ts_with_exog"),