From 6a44331f62dc524c907e3413536abd13622d7775 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 3 May 2023 11:00:08 +0300
Subject: [PATCH 01/18] save exog last available date

---
 etna/transforms/base.py                      | 18 ++++++++++++++++
 tests/test_transforms/test_base/test_base.py | 22 ++++++++++++++++++++
 2 files changed, 40 insertions(+)

diff --git a/etna/transforms/base.py b/etna/transforms/base.py
index bd2e03ee9..50ce0ac3f 100644
--- a/etna/transforms/base.py
+++ b/etna/transforms/base.py
@@ -26,6 +26,7 @@ class Transform(SaveMixin, AbstractSaveable, BaseMixin):
 
     def __init__(self, required_features: Union[Literal["all"], List[str]]):
         self.required_features = required_features
+        self._exog_last_date: Optional[Dict[str, pd.Timestamp]] = None
 
     @abstractmethod
     def get_regressors_info(self) -> List[str]:
@@ -87,6 +88,10 @@ def fit(self, ts: TSDataset) -> "Transform":
             The fitted transform instance.
         """
         df = ts.to_pandas(flatten=False, features=self.required_features)
+
+        if ts.df_exog is not None:
+            self._save_exog_last_date(df_exog=ts.df_exog)
+
         self._fit(df=df)
         return self
 
@@ -127,6 +132,19 @@ def transform(self, ts: TSDataset) -> TSDataset:
         ts = self._update_dataset(ts=ts, columns_before=columns_before, df_transformed=df_transformed)
         return ts
 
+    def _save_exog_last_date(self, df_exog: pd.DataFrame):
+        """Save last available date of each exogenous variable."""
+        exog_names = set(df_exog.columns.get_level_values("feature"))
+
+        self._exog_last_date = dict()
+        for name in exog_names:
+            feature = df_exog.loc[:, pd.IndexSlice[:, name]]
+
+            na_mask = pd.isna(feature).any(axis=1)
+            last_date = feature.index[~na_mask].max()
+
+            self._exog_last_date[name] = last_date
+
     def fit_transform(self, ts: TSDataset) -> TSDataset:
         """Fit and transform TSDataset.
 
diff --git a/tests/test_transforms/test_base/test_base.py b/tests/test_transforms/test_base/test_base.py
index 03cd58b5f..8f6c5ca1e 100644
--- a/tests/test_transforms/test_base/test_base.py
+++ b/tests/test_transforms/test_base/test_base.py
@@ -234,3 +234,25 @@ def test_inverse_transform_with_target_components_target_not_in_required_feature
     transform = AddConstTransform(in_column="exog", value=-10)
     transform.inverse_transform(ts=ts_with_target_components)
     pd.testing.assert_frame_equal(ts_with_target_components.get_target_components(), target_components_before)
+
+
+@pytest.fixture()
+def df_exog_with_nans():
+    df = pd.DataFrame(
+        {
+            "timestamp": list(pd.date_range("2023-01-01", periods=5)) * 2,
+            "segment": ["A"] * 5 + ["B"] * 5,
+            "feat1": [1, 2, 3, 4, None] + [1, 2, 3, 4, 5],
+            "feat2": [1, 2, 3, None, None] + [1, 2, 3, None, None],
+        }
+    )
+
+    return TSDataset.to_dataset(df=df)
+
+
+def test_save_exog_last_date(
+    df_exog_with_nans, expected={"feat1": pd.Timestamp("2023-01-04"), "feat2": pd.Timestamp("2023-01-03")}
+):
+    tr = TransformMock(required_features="all")
+    tr._save_exog_last_date(df_exog=df_exog_with_nans)
+    assert tr._exog_last_date == expected

From df3f3dabc4cf029555cc48b5f2dce3d81f7957bb Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 3 May 2023 11:00:44 +0300
Subject: [PATCH 02/18] added exog shift transform

---
 etna/transforms/__init__.py      |   1 +
 etna/transforms/math/__init__.py |   1 +
 etna/transforms/math/lags.py     | 147 +++++++++++++++++++++++++++++++
 3 files changed, 149 insertions(+)

diff --git a/etna/transforms/__init__.py b/etna/transforms/__init__.py
index 00c67b6ce..4c8332cae 100644
--- a/etna/transforms/__init__.py
+++ b/etna/transforms/__init__.py
@@ -26,6 +26,7 @@
 from etna.transforms.math import AddConstTransform
 from etna.transforms.math import BoxCoxTransform
 from etna.transforms.math import DifferencingTransform
+from etna.transforms.math import ExogShiftTransform
 from etna.transforms.math import LagTransform
 from etna.transforms.math import LambdaTransform
 from etna.transforms.math import LogTransform
diff --git a/etna/transforms/math/__init__.py b/etna/transforms/math/__init__.py
index 71877775d..dc3df9b09 100644
--- a/etna/transforms/math/__init__.py
+++ b/etna/transforms/math/__init__.py
@@ -1,6 +1,7 @@
 from etna.transforms.math.add_constant import AddConstTransform
 from etna.transforms.math.apply_lambda import LambdaTransform
 from etna.transforms.math.differencing import DifferencingTransform
+from etna.transforms.math.lags import ExogShiftTransform
 from etna.transforms.math.lags import LagTransform
 from etna.transforms.math.log import LogTransform
 from etna.transforms.math.power import BoxCoxTransform
diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index aae433624..7db80a8fe 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -1,4 +1,6 @@
+from typing import Dict
 from typing import List
+from typing import Literal
 from typing import Optional
 from typing import Union
 
@@ -96,3 +98,148 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
     def get_regressors_info(self) -> List[str]:
         """Return the list with regressors created by the transform."""
         return [self._get_column_name(lag) for lag in self.lags]
+
+
+class ExogShiftTransform(IrreversibleTransform, FutureMixin):
+    """Shifts exogenous variables from a given dataframe."""
+
+    def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = None):
+        """Create instance of ExogShiftTransform.
+
+        Parameters
+        ----------
+        lag:
+            value for shift estimation
+
+            * if set to `int` all exogenous variables will be shifted `lag` steps forward;
+
+            * if set to `auto` minimal shift will be estimated for each variable based on
+              the prediction horizon and available timeline
+
+        horizon:
+            prediction horizon. Mandatory when set to `lag="auto"`, ignored otherwise
+        """
+        super().__init__(required_features="all")
+
+        self.lag: Optional[int] = None
+        self.horizon: Optional[int] = None
+        self.auto = False
+
+        self._created_regressors: Optional[List[str]] = None
+        self._exog_shifts: Optional[Dict[str, int]] = None
+
+        if isinstance(lag, int):
+            if lag <= 0:
+                raise ValueError(f"{type(self).__name__} works only with positive lags values, {lag} given")
+            self.lag = lag
+
+        else:
+            if horizon is None:
+                raise ValueError("`horizon` should be specified when using `auto`!")
+
+            if horizon < 1:
+                raise ValueError(f"{type(self).__name__} works only with positive horizon values, {horizon} given")
+
+            self.horizon = horizon
+            self.auto = True
+
+    def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
+        """Estimate shifts for exogenous variables.
+
+        Parameters
+        ----------
+        df:
+            dataframe with data.
+
+        Returns
+        -------
+        result: ExogShiftTransform
+        """
+        feature_names = self._get_feature_names(df=df)
+
+        self._exog_shifts = dict()
+        self._created_regressors = []
+
+        for feature_name in feature_names:
+            shift = self._estimate_shift(df=df, feature_name=feature_name)
+            self._exog_shifts[feature_name] = shift
+
+            if shift > 0:
+                self._created_regressors.append(f"{feature_name}_shift_{shift}")
+
+        return self
+
+    @staticmethod
+    def _get_feature_names(df: pd.DataFrame) -> List[str]:
+        """Return the names of exogenous variables."""
+        names = set(df.columns.get_level_values("feature"))
+        if "target" in names:
+            names.remove("target")
+
+        return list(names)
+
+    def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
+        """Estimate shift value for exogenous variable."""
+        if not self.auto:
+            return self.lag  # type: ignore
+
+        freq = pd.infer_freq(df.index)
+
+        last_date = df.index.max() + pd.Timedelta(self.horizon, unit=freq)
+        last_feature_date = self._exog_last_date[feature_name]  # type: ignore
+
+        delta = last_date - last_feature_date
+        shift = max(0, int(delta / pd.Timedelta(1, unit=freq)))
+
+        return shift
+
+    def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
+        """Shift exogenous variables.
+
+        Parameters
+        ----------
+        df:
+            dataframe with data to transform.
+
+        Returns
+        -------
+        result: pd.Dataframe
+            transformed dataframe
+        """
+        if self._exog_shifts is None:
+            raise ValueError("Transform is not fitted!")
+
+        result = df
+        freq = pd.infer_freq(df.index)
+        segments = sorted(set(df.columns.get_level_values("segment")))
+        feature_names = self._get_feature_names(df=df)
+
+        shifted_features = []
+        features_to_remove = []
+        for feature_name in feature_names:
+            shift = self._exog_shifts[feature_name]
+
+            feature = df.loc[:, pd.IndexSlice[:, feature_name]]
+
+            if shift > 0:
+                shifted_feature = feature.shift(shift, freq=freq)
+
+                column_name = f"{feature_name}_shift_{shift}"
+                shifted_feature.columns = pd.MultiIndex.from_product([segments, [column_name]])
+
+                shifted_features.append(shifted_feature)
+                features_to_remove.append(feature_name)
+
+        if len(features_to_remove) > 0:
+            result = result.drop(columns=pd.MultiIndex.from_product([segments, features_to_remove]))
+
+        result = pd.concat([result] + shifted_features, axis=1)
+        result.sort_index(axis=1, inplace=True)
+        return result
+
+    def get_regressors_info(self) -> List[str]:
+        """Return the list with regressors created by the transform."""
+        if self._created_regressors is None:
+            return []
+
+        return self._created_regressors

From 01dc9b00b17ee22e752a90c01b5947105d9e655b Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 3 May 2023 11:00:57 +0300
Subject: [PATCH 03/18] added tests

---
 .../test_math/test_exog_shift_transform.py    | 131 ++++++++++++++++++
 1 file changed, 131 insertions(+)
 create mode 100644 tests/test_transforms/test_math/test_exog_shift_transform.py

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
new file mode 100644
index 000000000..a11c44240
--- /dev/null
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -0,0 +1,131 @@
+import numpy as np
+import pandas as pd
+import pytest
+
+from etna.datasets import TSDataset
+from etna.metrics import MAE
+from etna.models import LinearPerSegmentModel
+from etna.pipeline import Pipeline
+from etna.transforms import ExogShiftTransform
+
+
+@pytest.fixture()
+def df_exog_with_nans():
+    df = pd.DataFrame(
+        {
+            "timestamp": list(pd.date_range("2023-01-01", periods=5)) * 2,
+            "segment": ["A"] * 5 + ["B"] * 5,
+            "feat1": [1, 2, 3, 4, None] + [1, 2, 3, 4, 5],
+            "feat2": [1, 2, 3, None, None] + [1, 2, 3, None, None],
+            "feat3": [1, 2, 3, 4, 5] + [1, 2, 3, 4, 5],
+        }
+    )
+
+    return TSDataset.to_dataset(df=df)
+
+
+@pytest.fixture()
+def ts_with_exogs(df_exog_with_nans):
+    df = pd.DataFrame(
+        {
+            "timestamp": list(pd.date_range("2023-01-01", periods=4)) * 2,
+            "segment": ["A"] * 4 + ["B"] * 4,
+            "target": list(3 * np.arange(1, 5)) * 2,
+        }
+    )
+
+    df = TSDataset.to_dataset(df=df)
+    ts = TSDataset(df=df, df_exog=df_exog_with_nans, freq="D")
+    return ts
+
+
+def test_negative_lag():
+    with pytest.raises(ValueError, match=".* works only with positive lags"):
+        ExogShiftTransform(lag=-1)
+
+
+def test_horizon_not_set():
+    with pytest.raises(ValueError, match="`horizon` should be specified"):
+        ExogShiftTransform(lag="auto")
+
+
+def test_negative_horizon_set():
+    with pytest.raises(ValueError, match=".* works only with positive horizon"):
+        ExogShiftTransform(lag="auto", horizon=-1)
+
+
+def test_get_feature_names(example_reg_tsds, expected={"regressor_exog_weekend"}):
+    feature_names = ExogShiftTransform._get_feature_names(example_reg_tsds.df)
+    assert set(feature_names) == expected
+
+
+def test_regressors_info_not_fit():
+    assert ExogShiftTransform(lag=1).get_regressors_info() == []
+
+
+@pytest.mark.parametrize(
+    "horizon,expected",
+    ((1, {"feat1_shift_1", "feat2_shift_2"}), (2, {"feat1_shift_2", "feat2_shift_3", "feat3_shift_1"})),
+)
+def test_regressors_info(ts_with_exogs, horizon, expected):
+    t = ExogShiftTransform(lag="auto", horizon=horizon)
+    t.fit(ts=ts_with_exogs)
+    assert set(t.get_regressors_info()) == expected
+
+
+@pytest.mark.parametrize(
+    "lag,horizon,expected",
+    (
+        (1, None, {"feat1": 1, "feat2": 1, "feat3": 1}),
+        ("auto", 1, {"feat1": 1, "feat2": 2, "feat3": 0}),
+        ("auto", 2, {"feat1": 2, "feat2": 3, "feat3": 1}),
+    ),
+)
+def test_estimate_shift(ts_with_exogs, lag, horizon, expected):
+    t = ExogShiftTransform(lag=lag, horizon=horizon)
+    t.fit(ts=ts_with_exogs)
+    assert t._exog_shifts == expected
+
+
+@pytest.mark.parametrize("lag", (1, "auto"))
+def test_shift_no_exog(simple_df, lag, expected={"target"}):
+    t = ExogShiftTransform(lag=lag, horizon=1)
+    transformed = t.fit_transform(simple_df)
+    assert set(transformed.df.columns.get_level_values("feature")) == expected
+
+
+@pytest.mark.parametrize(
+    "lag,horizon,expected",
+    (
+        (1, None, {"feat1_shift_1", "feat2_shift_1", "feat3_shift_1", "target"}),
+        ("auto", 1, {"feat1_shift_1", "feat2_shift_2", "feat3", "target"}),
+        ("auto", 2, {"feat1_shift_2", "feat2_shift_3", "feat3_shift_1", "target"}),
+    ),
+)
+def test_transformed_names(ts_with_exogs, lag, horizon, expected):
+    t = ExogShiftTransform(lag=lag, horizon=horizon)
+    transformed = t.fit_transform(ts=ts_with_exogs)
+    column_names = transformed.df.columns.get_level_values("feature")
+    assert set(column_names) == expected
+
+
+@pytest.mark.parametrize("lag", (3, "auto"))
+@pytest.mark.parametrize("horizon", range(1, 3))
+def test_pipeline_forecast(ts_with_exogs, lag, horizon):
+    pipeline = Pipeline(
+        transforms=[ExogShiftTransform(lag=lag, horizon=horizon)], model=LinearPerSegmentModel(), horizon=horizon
+    )
+    pipeline.fit(ts_with_exogs)
+    pipeline.forecast()
+
+
+@pytest.mark.parametrize("lag", (3, "auto"))
+@pytest.mark.parametrize("horizon", range(7, 10))
+def test_pipeline_backtest(example_reg_tsds, lag, horizon):
+    ts = example_reg_tsds
+
+    pipeline = Pipeline(
+        transforms=[ExogShiftTransform(lag=lag, horizon=horizon)], model=LinearPerSegmentModel(), horizon=horizon
+    )
+
+    pipeline.backtest(ts=ts, metrics=[MAE()])

From 6b42654dd14166c2b043c91f11fb18bb438692ce Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 3 May 2023 11:57:53 +0300
Subject: [PATCH 04/18] fixed tests

---
 etna/transforms/base.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/etna/transforms/base.py b/etna/transforms/base.py
index 50ce0ac3f..8e6962b86 100644
--- a/etna/transforms/base.py
+++ b/etna/transforms/base.py
@@ -89,8 +89,9 @@ def fit(self, ts: TSDataset) -> "Transform":
         """
         df = ts.to_pandas(flatten=False, features=self.required_features)
 
-        if ts.df_exog is not None:
-            self._save_exog_last_date(df_exog=ts.df_exog)
+        df_exog = ts.df_exog
+        if df_exog is not None and isinstance(df_exog, pd.DataFrame):
+            self._save_exog_last_date(df_exog=df_exog)
 
         self._fit(df=df)
         return self

From d8fa4c25906b1f4491c1d5bdab3afdf79fd0a740 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 12:09:05 +0300
Subject: [PATCH 05/18] fixed feature names selection

---
 etna/transforms/math/lags.py | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index 7db80a8fe..1f696657e 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -173,10 +173,13 @@ def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
     def _get_feature_names(df: pd.DataFrame) -> List[str]:
         """Return the names of exogenous variables."""
         names = set(df.columns.get_level_values("feature"))
-        if "target" in names:
-            names.remove("target")
 
-        return list(names)
+        names_to_remove = {"target"}
+        names_to_remove |= match_target_quantiles(names)
+        names_to_remove |= match_target_components(names)
+
+        features = names - names_to_remove
+        return list(features)
 
     def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
         """Estimate shift value for exogenous variable."""

From 7070b1838b3c2d5044e3f114f47b79492e5470ef Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 12:09:52 +0300
Subject: [PATCH 06/18] new shift estimation logic

---
 etna/transforms/math/lags.py | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index 1f696657e..dfe9b8dd0 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -188,11 +188,19 @@ def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
 
         freq = pd.infer_freq(df.index)
 
-        last_date = df.index.max() + pd.Timedelta(self.horizon, unit=freq)
+        last_date = df.index.max()
         last_feature_date = self._exog_last_date[feature_name]  # type: ignore
 
-        delta = last_date - last_feature_date
-        shift = max(0, int(delta / pd.Timedelta(1, unit=freq)))
+        if last_feature_date > last_date:
+            delta = -determine_num_steps(start_timestamp=last_date, end_timestamp=last_feature_date, freq=freq)
+
+        elif last_feature_date < last_date:
+            delta = determine_num_steps(start_timestamp=last_feature_date, end_timestamp=last_date, freq=freq)
+
+        else:
+            delta = 0
+
+        shift = max(0, delta + self.horizon)  # type: ignore
 
         return shift
 

From de77c595e1d72b5a45242b25b73c765856417b2d Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 12:10:27 +0300
Subject: [PATCH 07/18] review fixes

---
 etna/transforms/math/lags.py | 24 ++++++++++++++----------
 1 file changed, 14 insertions(+), 10 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index dfe9b8dd0..da40ff496 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -6,6 +6,9 @@
 
 import pandas as pd
 
+from etna.datasets.utils import match_target_components
+from etna.datasets.utils import match_target_quantiles
+from etna.models.utils import determine_num_steps
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversibleTransform
 
@@ -123,25 +126,25 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
 
         self.lag: Optional[int] = None
         self.horizon: Optional[int] = None
-        self.auto = False
+        self._auto = False
 
         self._created_regressors: Optional[List[str]] = None
         self._exog_shifts: Optional[Dict[str, int]] = None
 
         if isinstance(lag, int):
             if lag <= 0:
-                raise ValueError(f"{type(self).__name__} works only with positive lags values, {lag} given")
+                raise ValueError(f"{self.__class__.__name__} works only with positive lags values, {lag} given")
             self.lag = lag
 
         else:
             if horizon is None:
-                raise ValueError("`horizon` should be specified when using `auto`!")
+                raise ValueError("Value of `horizon` should be specified when using `auto`!")
 
             if horizon < 1:
-                raise ValueError(f"{type(self).__name__} works only with positive horizon values, {horizon} given")
+                raise ValueError(f"{self.__class__.__name__} works only with positive horizon values, {horizon} given")
 
             self.horizon = horizon
-            self.auto = True
+            self._auto = True
 
     def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
         """Estimate shifts for exogenous variables.
@@ -153,7 +156,8 @@ def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
 
         Returns
         -------
-        result: ExogShiftTransform
+        :
+            Fitted `ExogShiftTransform` instance.
         """
         feature_names = self._get_feature_names(df=df)
 
@@ -183,7 +187,7 @@ def _get_feature_names(df: pd.DataFrame) -> List[str]:
 
     def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
         """Estimate shift value for exogenous variable."""
-        if not self.auto:
+        if not self._auto:
             return self.lag  # type: ignore
 
         freq = pd.infer_freq(df.index)
@@ -214,8 +218,8 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
 
         Returns
         -------
-        result: pd.Dataframe
-            transformed dataframe
+        :
+            Transformed dataframe.
         """
         if self._exog_shifts is None:
             raise ValueError("Transform is not fitted!")
@@ -251,6 +255,6 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
     def get_regressors_info(self) -> List[str]:
         """Return the list with regressors created by the transform."""
         if self._created_regressors is None:
-            return []
+            raise ValueError("Fit the transform to get the regressors info!")
 
         return self._created_regressors

From 70a7d5f9614793e3e3c9ac5f53aade343e15357a Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 12:11:22 +0300
Subject: [PATCH 08/18] added tests

---
 .../test_math/test_exog_shift_transform.py        | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index a11c44240..93d4d02b5 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -54,13 +54,24 @@ def test_negative_horizon_set():
         ExogShiftTransform(lag="auto", horizon=-1)
 
 
+def test_regressors_info_not_fit():
+    with pytest.raises(ValueError, match="Fit the transform"):
+        ExogShiftTransform(lag=1).get_regressors_info()
+
+
 def test_get_feature_names(example_reg_tsds, expected={"regressor_exog_weekend"}):
     feature_names = ExogShiftTransform._get_feature_names(example_reg_tsds.df)
     assert set(feature_names) == expected
 
 
-def test_regressors_info_not_fit():
-    assert ExogShiftTransform(lag=1).get_regressors_info() == []
+@pytest.mark.parametrize(
+    "ts_name", ("toy_dataset_with_mean_shift_in_target", "product_level_constant_forecast_with_target_components")
+)
+def test_ts_with_quantiles_and_components(ts_name, request):
+    ts = request.getfixturevalue(ts_name)
+    t = ExogShiftTransform(lag=1)
+    t.fit(ts=ts)
+    assert t.get_regressors_info() == []
 
 
 @pytest.mark.parametrize(

From bdf1b6983a0025a359758cce9fb0c238d5c7a462 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 12:11:36 +0300
Subject: [PATCH 09/18] reworked tests

---
 .../test_math/test_exog_shift_transform.py    | 48 ++++++++++++++-----
 1 file changed, 37 insertions(+), 11 deletions(-)

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index 93d4d02b5..400b08eb1 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -9,11 +9,10 @@
 from etna.transforms import ExogShiftTransform
 
 
-@pytest.fixture()
-def df_exog_with_nans():
+def build_df_exog_with_nans(freq="D"):
     df = pd.DataFrame(
         {
-            "timestamp": list(pd.date_range("2023-01-01", periods=5)) * 2,
+            "timestamp": list(pd.date_range("2023-01-01", periods=5, freq=freq)) * 2,
             "segment": ["A"] * 5 + ["B"] * 5,
             "feat1": [1, 2, 3, 4, None] + [1, 2, 3, 4, 5],
             "feat2": [1, 2, 3, None, None] + [1, 2, 3, None, None],
@@ -24,21 +23,30 @@ def df_exog_with_nans():
     return TSDataset.to_dataset(df=df)
 
 
-@pytest.fixture()
-def ts_with_exogs(df_exog_with_nans):
+def build_ts_with_exogs(freq="D"):
     df = pd.DataFrame(
         {
-            "timestamp": list(pd.date_range("2023-01-01", periods=4)) * 2,
+            "timestamp": list(pd.date_range("2023-01-01", periods=4, freq=freq)) * 2,
             "segment": ["A"] * 4 + ["B"] * 4,
             "target": list(3 * np.arange(1, 5)) * 2,
         }
     )
 
     df = TSDataset.to_dataset(df=df)
-    ts = TSDataset(df=df, df_exog=df_exog_with_nans, freq="D")
+    ts = TSDataset(df=df, df_exog=build_df_exog_with_nans(freq=freq), freq=freq)
     return ts
 
 
+@pytest.fixture()
+def ts_with_exogs():
+    return build_ts_with_exogs(freq="D")
+
+
+@pytest.fixture()
+def ts_with_exogs_ms_freq():
+    return build_ts_with_exogs(freq="MS")
+
+
 def test_negative_lag():
     with pytest.raises(ValueError, match=".* works only with positive lags"):
         ExogShiftTransform(lag=-1)
@@ -113,20 +121,38 @@ def test_shift_no_exog(simple_df, lag, expected={"target"}):
         ("auto", 2, {"feat1_shift_2", "feat2_shift_3", "feat3_shift_1", "target"}),
     ),
 )
-def test_transformed_names(ts_with_exogs, lag, horizon, expected):
+@pytest.mark.parametrize(
+    "ts_name",
+    (
+        "ts_with_exogs",
+        "ts_with_exogs_ms_freq",
+    ),
+)
+def test_transformed_names(ts_name, lag, horizon, expected, request):
+    ts = request.getfixturevalue(ts_name)
+
     t = ExogShiftTransform(lag=lag, horizon=horizon)
-    transformed = t.fit_transform(ts=ts_with_exogs)
+    transformed = t.fit_transform(ts=ts)
     column_names = transformed.df.columns.get_level_values("feature")
     assert set(column_names) == expected
 
 
 @pytest.mark.parametrize("lag", (3, "auto"))
 @pytest.mark.parametrize("horizon", range(1, 3))
-def test_pipeline_forecast(ts_with_exogs, lag, horizon):
+@pytest.mark.parametrize(
+    "ts_name",
+    (
+        "ts_with_exogs",
+        "ts_with_exogs_ms_freq",
+    ),
+)
+def test_pipeline_forecast(ts_name, lag, horizon, request):
+    ts = request.getfixturevalue(ts_name)
+
     pipeline = Pipeline(
         transforms=[ExogShiftTransform(lag=lag, horizon=horizon)], model=LinearPerSegmentModel(), horizon=horizon
     )
-    pipeline.fit(ts_with_exogs)
+    pipeline.fit(ts)
     pipeline.forecast()
 
 

From 3a1a3d95fc302685e55f8ef5ba3e5d8e8339c350 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 14:10:07 +0300
Subject: [PATCH 10/18] moved exog dates from base

---
 etna/transforms/base.py                       | 19 ----------
 etna/transforms/math/lags.py                  | 37 +++++++++++++++++++
 tests/test_transforms/test_base/test_base.py  | 22 -----------
 .../test_math/test_exog_shift_transform.py    | 20 ++++++++++
 4 files changed, 57 insertions(+), 41 deletions(-)

diff --git a/etna/transforms/base.py b/etna/transforms/base.py
index 8e6962b86..bd2e03ee9 100644
--- a/etna/transforms/base.py
+++ b/etna/transforms/base.py
@@ -26,7 +26,6 @@ class Transform(SaveMixin, AbstractSaveable, BaseMixin):
 
     def __init__(self, required_features: Union[Literal["all"], List[str]]):
         self.required_features = required_features
-        self._exog_last_date: Optional[Dict[str, pd.Timestamp]] = None
 
     @abstractmethod
     def get_regressors_info(self) -> List[str]:
@@ -88,11 +87,6 @@ def fit(self, ts: TSDataset) -> "Transform":
             The fitted transform instance.
         """
         df = ts.to_pandas(flatten=False, features=self.required_features)
-
-        df_exog = ts.df_exog
-        if df_exog is not None and isinstance(df_exog, pd.DataFrame):
-            self._save_exog_last_date(df_exog=df_exog)
-
         self._fit(df=df)
         return self
 
@@ -133,19 +127,6 @@ def transform(self, ts: TSDataset) -> TSDataset:
         ts = self._update_dataset(ts=ts, columns_before=columns_before, df_transformed=df_transformed)
         return ts
 
-    def _save_exog_last_date(self, df_exog: pd.DataFrame):
-        """Save last available date of each exogenous variable."""
-        exog_names = set(df_exog.columns.get_level_values("feature"))
-
-        self._exog_last_date = dict()
-        for name in exog_names:
-            feature = df_exog.loc[:, pd.IndexSlice[:, name]]
-
-            na_mask = pd.isna(feature).any(axis=1)
-            last_date = feature.index[~na_mask].max()
-
-            self._exog_last_date[name] = last_date
-
     def fit_transform(self, ts: TSDataset) -> TSDataset:
         """Fit and transform TSDataset.
 
diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index da40ff496..cc38f3ad6 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -6,6 +6,7 @@
 
 import pandas as pd
 
+from etna.datasets import TSDataset
 from etna.datasets.utils import match_target_components
 from etna.datasets.utils import match_target_quantiles
 from etna.models.utils import determine_num_steps
@@ -130,6 +131,7 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
 
         self._created_regressors: Optional[List[str]] = None
         self._exog_shifts: Optional[Dict[str, int]] = None
+        self._exog_last_date: Optional[Dict[str, pd.Timestamp]] = None
 
         if isinstance(lag, int):
             if lag <= 0:
@@ -146,6 +148,41 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
             self.horizon = horizon
             self._auto = True
 
+    def _save_exog_last_date(self, df_exog: pd.DataFrame):
+        """Save last available date of each exogenous variable."""
+        exog_names = set(df_exog.columns.get_level_values("feature"))
+
+        self._exog_last_date = dict()
+        for name in exog_names:
+            feature = df_exog.loc[:, pd.IndexSlice[:, name]]
+
+            na_mask = pd.isna(feature).any(axis=1)
+            last_date = feature.index[~na_mask].max()
+
+            self._exog_last_date[name] = last_date
+
+    def fit(self, ts: TSDataset) -> "ExogShiftTransform":
+        """Fit the transform.
+
+        Parameters
+        ----------
+        ts:
+            Dataset to fit the transform on.
+
+        Returns
+        -------
+        :
+            The fitted transform instance.
+        """
+
+        df_exog = ts.df_exog
+        if df_exog is not None and isinstance(df_exog, pd.DataFrame):
+            self._save_exog_last_date(df_exog=df_exog)
+
+        super().fit(ts=ts)
+
+        return self
+
     def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
         """Estimate shifts for exogenous variables.
 
diff --git a/tests/test_transforms/test_base/test_base.py b/tests/test_transforms/test_base/test_base.py
index 8f6c5ca1e..03cd58b5f 100644
--- a/tests/test_transforms/test_base/test_base.py
+++ b/tests/test_transforms/test_base/test_base.py
@@ -234,25 +234,3 @@ def test_inverse_transform_with_target_components_target_not_in_required_feature
     transform = AddConstTransform(in_column="exog", value=-10)
     transform.inverse_transform(ts=ts_with_target_components)
     pd.testing.assert_frame_equal(ts_with_target_components.get_target_components(), target_components_before)
-
-
-@pytest.fixture()
-def df_exog_with_nans():
-    df = pd.DataFrame(
-        {
-            "timestamp": list(pd.date_range("2023-01-01", periods=5)) * 2,
-            "segment": ["A"] * 5 + ["B"] * 5,
-            "feat1": [1, 2, 3, 4, None] + [1, 2, 3, 4, 5],
-            "feat2": [1, 2, 3, None, None] + [1, 2, 3, None, None],
-        }
-    )
-
-    return TSDataset.to_dataset(df=df)
-
-
-def test_save_exog_last_date(
-    df_exog_with_nans, expected={"feat1": pd.Timestamp("2023-01-04"), "feat2": pd.Timestamp("2023-01-03")}
-):
-    tr = TransformMock(required_features="all")
-    tr._save_exog_last_date(df_exog=df_exog_with_nans)
-    assert tr._exog_last_date == expected
diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index 400b08eb1..bfa9ca23e 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -37,6 +37,11 @@ def build_ts_with_exogs(freq="D"):
     return ts
 
 
+@pytest.fixture()
+def df_exog_with_nans():
+    return build_df_exog_with_nans(freq="D")
+
+
 @pytest.fixture()
 def ts_with_exogs():
     return build_ts_with_exogs(freq="D")
@@ -46,6 +51,21 @@ def ts_with_exogs():
 def ts_with_exogs_ms_freq():
     return build_ts_with_exogs(freq="MS")
 
+@pytest.mark.parametrize(
+    "expected",
+    (
+            {
+                "feat1": pd.Timestamp("2023-01-04"),
+                "feat2": pd.Timestamp("2023-01-03"),
+                'feat3': pd.Timestamp('2023-01-05 00:00:00')
+            },
+    )
+)
+def test_save_exog_last_date(df_exog_with_nans, expected):
+    t = ExogShiftTransform(lag=1)
+    t._save_exog_last_date(df_exog=df_exog_with_nans)
+    assert t._exog_last_date == expected
+
 
 def test_negative_lag():
     with pytest.raises(ValueError, match=".* works only with positive lags"):

From e62c4c9fcd0e5d2e6b08076de6542f75d2a205f9 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 15:03:10 +0300
Subject: [PATCH 11/18] reworked `_get_feature_names`

---
 etna/transforms/math/lags.py | 22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index cc38f3ad6..ec0c082bf 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -7,8 +7,6 @@
 import pandas as pd
 
 from etna.datasets import TSDataset
-from etna.datasets.utils import match_target_components
-from etna.datasets.utils import match_target_quantiles
 from etna.models.utils import determine_num_steps
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversibleTransform
@@ -132,6 +130,7 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
         self._created_regressors: Optional[List[str]] = None
         self._exog_shifts: Optional[Dict[str, int]] = None
         self._exog_last_date: Optional[Dict[str, pd.Timestamp]] = None
+        self._filter_out_columns = {"target"}
 
         if isinstance(lag, int):
             if lag <= 0:
@@ -210,17 +209,20 @@ def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
 
         return self
 
-    @staticmethod
-    def _get_feature_names(df: pd.DataFrame) -> List[str]:
+    def _get_feature_names(self, df: pd.DataFrame) -> List[str]:
         """Return the names of exogenous variables."""
-        names = set(df.columns.get_level_values("feature"))
+        if self._exog_last_date is not None:
+            feature_names = list(self._exog_last_date.keys())
 
-        names_to_remove = {"target"}
-        names_to_remove |= match_target_quantiles(names)
-        names_to_remove |= match_target_components(names)
+        else:
+            feature_names = []
+
+        df_columns = df.columns.get_level_values("feature")
+        for name in feature_names:
+            if name not in df_columns:
+                raise ValueError(f"Feature `{name}` is expected to be in the dataframe!")
 
-        features = names - names_to_remove
-        return list(features)
+        return feature_names
 
     def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
         """Estimate shift value for exogenous variable."""

From 6e6572e7a876efac67ac65501d52f67a0ffa736b Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 15:03:33 +0300
Subject: [PATCH 12/18] added tests

---
 .../test_math/test_exog_shift_transform.py          | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index bfa9ca23e..efc07484f 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -88,10 +88,21 @@ def test_regressors_info_not_fit():
 
 
 def test_get_feature_names(example_reg_tsds, expected={"regressor_exog_weekend"}):
-    feature_names = ExogShiftTransform._get_feature_names(example_reg_tsds.df)
+    t = ExogShiftTransform(lag=1)
+    t.fit(ts=example_reg_tsds)
+    feature_names = t._get_feature_names(example_reg_tsds.df)
     assert set(feature_names) == expected
 
 
+def test_get_feature_names_no_exog_error(ts_with_exogs):
+    t = ExogShiftTransform(lag=1)
+    t.fit(ts=ts_with_exogs)
+    df = ts_with_exogs.df.drop(columns=["feat3"], level=1)
+
+    with pytest.raises(ValueError, match="Feature `feat3` is expected to be in the dataframe!"):
+        t._get_feature_names(df)
+
+
 @pytest.mark.parametrize(
     "ts_name", ("toy_dataset_with_mean_shift_in_target", "product_level_constant_forecast_with_target_components")
 )

From 826ed0e424db47882d3babcac81f11b70a16bda9 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Fri, 5 May 2023 15:42:52 +0300
Subject: [PATCH 13/18] formatting

---
 etna/transforms/math/lags.py                        |  1 -
 .../test_math/test_exog_shift_transform.py          | 13 +++++++------
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index ec0c082bf..023b10f23 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -173,7 +173,6 @@ def fit(self, ts: TSDataset) -> "ExogShiftTransform":
         :
             The fitted transform instance.
         """
-
         df_exog = ts.df_exog
         if df_exog is not None and isinstance(df_exog, pd.DataFrame):
             self._save_exog_last_date(df_exog=df_exog)
diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index efc07484f..317b5869b 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -51,15 +51,16 @@ def ts_with_exogs():
 def ts_with_exogs_ms_freq():
     return build_ts_with_exogs(freq="MS")
 
+
 @pytest.mark.parametrize(
     "expected",
     (
-            {
-                "feat1": pd.Timestamp("2023-01-04"),
-                "feat2": pd.Timestamp("2023-01-03"),
-                'feat3': pd.Timestamp('2023-01-05 00:00:00')
-            },
-    )
+        {
+            "feat1": pd.Timestamp("2023-01-04"),
+            "feat2": pd.Timestamp("2023-01-03"),
+            "feat3": pd.Timestamp("2023-01-05 00:00:00"),
+        },
+    ),
 )
 def test_save_exog_last_date(df_exog_with_nans, expected):
     t = ExogShiftTransform(lag=1)

From bced70bb18b82343c64fc7edb987cee9c3560e17 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 10 May 2023 11:39:18 +0300
Subject: [PATCH 14/18] review fixes

---
 CHANGELOG.md                 |  2 +-
 etna/transforms/math/lags.py | 22 ++++++++++++++--------
 2 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6b9b19610..fd8b07b0a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,7 +8,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased
 ### Added
 - Notebook `forecast_interpretation.ipynb` with forecast decomposition ([#1220](https://github.com/tinkoff-ai/etna/pull/1220))
--
+- Exogenous variables shift transform `ExogShiftTransform`([#1254](https://github.com/tinkoff-ai/etna/pull/1254))
 -
 - 
 ### Changed
diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index 023b10f23..e9397faef 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -127,6 +127,7 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
         self.horizon: Optional[int] = None
         self._auto = False
 
+        self._freq: Optional[str] = None
         self._created_regressors: Optional[List[str]] = None
         self._exog_shifts: Optional[Dict[str, int]] = None
         self._exog_last_date: Optional[Dict[str, pd.Timestamp]] = None
@@ -173,10 +174,15 @@ def fit(self, ts: TSDataset) -> "ExogShiftTransform":
         :
             The fitted transform instance.
         """
+        self._freq = ts.freq
         df_exog = ts.df_exog
-        if df_exog is not None and isinstance(df_exog, pd.DataFrame):
+
+        if df_exog is not None:
             self._save_exog_last_date(df_exog=df_exog)
 
+        else:
+            self._exog_last_date = {}
+
         super().fit(ts=ts)
 
         return self
@@ -210,7 +216,7 @@ def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
 
     def _get_feature_names(self, df: pd.DataFrame) -> List[str]:
         """Return the names of exogenous variables."""
-        if self._exog_last_date is not None:
+        if self._exog_last_date is not None and len(self._exog_last_date) > 0:
             feature_names = list(self._exog_last_date.keys())
 
         else:
@@ -228,16 +234,17 @@ def _estimate_shift(self, df: pd.DataFrame, feature_name: str) -> int:
         if not self._auto:
             return self.lag  # type: ignore
 
-        freq = pd.infer_freq(df.index)
+        if self._exog_last_date is None or self._freq is None:
+            raise ValueError("Call `fit()` method before estimating exog shifts!")
 
         last_date = df.index.max()
-        last_feature_date = self._exog_last_date[feature_name]  # type: ignore
+        last_feature_date = self._exog_last_date[feature_name]
 
         if last_feature_date > last_date:
-            delta = -determine_num_steps(start_timestamp=last_date, end_timestamp=last_feature_date, freq=freq)
+            delta = -determine_num_steps(start_timestamp=last_date, end_timestamp=last_feature_date, freq=self._freq)
 
         elif last_feature_date < last_date:
-            delta = determine_num_steps(start_timestamp=last_feature_date, end_timestamp=last_date, freq=freq)
+            delta = determine_num_steps(start_timestamp=last_feature_date, end_timestamp=last_date, freq=self._freq)
 
         else:
             delta = 0
@@ -263,7 +270,6 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
             raise ValueError("Transform is not fitted!")
 
         result = df
-        freq = pd.infer_freq(df.index)
         segments = sorted(set(df.columns.get_level_values("segment")))
         feature_names = self._get_feature_names(df=df)
 
@@ -275,7 +281,7 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
             feature = df.loc[:, pd.IndexSlice[:, feature_name]]
 
             if shift > 0:
-                shifted_feature = feature.shift(shift, freq=freq)
+                shifted_feature = feature.shift(shift, freq=self._freq)
 
                 column_name = f"{feature_name}_shift_{shift}"
                 shifted_feature.columns = pd.MultiIndex.from_product([segments, [column_name]])

From 362ec4bbb512285376a7b6b26063616a91dcdca1 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 10 May 2023 11:39:30 +0300
Subject: [PATCH 15/18] added test

---
 tests/test_transforms/test_math/test_exog_shift_transform.py | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index 317b5869b..6ce7082e0 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -88,6 +88,11 @@ def test_regressors_info_not_fit():
         ExogShiftTransform(lag=1).get_regressors_info()
 
 
+def test_estimate_shift_not_fit():
+    with pytest.raises(ValueError, match=".* method before estimating exog shifts!"):
+        ExogShiftTransform(lag="auto", horizon=1)._estimate_shift(None, None)
+
+
 def test_get_feature_names(example_reg_tsds, expected={"regressor_exog_weekend"}):
     t = ExogShiftTransform(lag=1)
     t.fit(ts=example_reg_tsds)

From 49ed970bd2410c6aef9a0552893390cea39594a7 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 10 May 2023 13:48:15 +0300
Subject: [PATCH 16/18] moved exog check

---
 etna/transforms/math/lags.py | 25 ++++++++++---------------
 1 file changed, 10 insertions(+), 15 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index e9397faef..de88b2af6 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -148,18 +148,19 @@ def __init__(self, lag: Union[int, Literal["auto"]], horizon: Optional[int] = No
             self.horizon = horizon
             self._auto = True
 
-    def _save_exog_last_date(self, df_exog: pd.DataFrame):
+    def _save_exog_last_date(self, df_exog: Optional[pd.DataFrame] = None):
         """Save last available date of each exogenous variable."""
-        exog_names = set(df_exog.columns.get_level_values("feature"))
+        self._exog_last_date = {}
+        if df_exog is not None:
+            exog_names = set(df_exog.columns.get_level_values("feature"))
 
-        self._exog_last_date = dict()
-        for name in exog_names:
-            feature = df_exog.loc[:, pd.IndexSlice[:, name]]
+            for name in exog_names:
+                feature = df_exog.loc[:, pd.IndexSlice[:, name]]
 
-            na_mask = pd.isna(feature).any(axis=1)
-            last_date = feature.index[~na_mask].max()
+                na_mask = pd.isna(feature).any(axis=1)
+                last_date = feature.index[~na_mask].max()
 
-            self._exog_last_date[name] = last_date
+                self._exog_last_date[name] = last_date
 
     def fit(self, ts: TSDataset) -> "ExogShiftTransform":
         """Fit the transform.
@@ -175,13 +176,7 @@ def fit(self, ts: TSDataset) -> "ExogShiftTransform":
             The fitted transform instance.
         """
         self._freq = ts.freq
-        df_exog = ts.df_exog
-
-        if df_exog is not None:
-            self._save_exog_last_date(df_exog=df_exog)
-
-        else:
-            self._exog_last_date = {}
+        self._save_exog_last_date(df_exog=ts.df_exog)
 
         super().fit(ts=ts)
 

From 4dd6333e7adbf0f6948fb64c2952c2f1c2aa70b5 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 10 May 2023 15:09:10 +0300
Subject: [PATCH 17/18] reworked check

---
 etna/transforms/math/lags.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/etna/transforms/math/lags.py b/etna/transforms/math/lags.py
index de88b2af6..38d8f4f40 100644
--- a/etna/transforms/math/lags.py
+++ b/etna/transforms/math/lags.py
@@ -211,12 +211,10 @@ def _fit(self, df: pd.DataFrame) -> "ExogShiftTransform":
 
     def _get_feature_names(self, df: pd.DataFrame) -> List[str]:
         """Return the names of exogenous variables."""
-        if self._exog_last_date is not None and len(self._exog_last_date) > 0:
+        feature_names = []
+        if self._exog_last_date is not None:
             feature_names = list(self._exog_last_date.keys())
 
-        else:
-            feature_names = []
-
         df_columns = df.columns.get_level_values("feature")
         for name in feature_names:
             if name not in df_columns:

From 380fb7f0398bef6b9390c21d9339a97c5fcfc646 Mon Sep 17 00:00:00 2001
From: brsnw250 <maxzherelo1@gmail.com>
Date: Wed, 10 May 2023 15:09:28 +0300
Subject: [PATCH 18/18] added test

---
 .../test_transforms/test_math/test_exog_shift_transform.py  | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/tests/test_transforms/test_math/test_exog_shift_transform.py b/tests/test_transforms/test_math/test_exog_shift_transform.py
index 6ce7082e0..3d13319fe 100644
--- a/tests/test_transforms/test_math/test_exog_shift_transform.py
+++ b/tests/test_transforms/test_math/test_exog_shift_transform.py
@@ -93,6 +93,12 @@ def test_estimate_shift_not_fit():
         ExogShiftTransform(lag="auto", horizon=1)._estimate_shift(None, None)
 
 
+def test_transform_not_fit(ts_with_exogs):
+    t = ExogShiftTransform(lag=1)
+    with pytest.raises(ValueError, match="Transform is not fitted!"):
+        t.transform(ts=ts_with_exogs)
+
+
 def test_get_feature_names(example_reg_tsds, expected={"regressor_exog_weekend"}):
     t = ExogShiftTransform(lag=1)
     t.fit(ts=example_reg_tsds)