Add default params_to_tune for DateFlagsTransform, TimeFlagsTransform, SpecialDaysTransform, FourierTransform

CHANGELOG.md

@@ -51,6 +51,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Add default `params_to_tune` for `DeepARModel` and `TFTModel` ([#1210](https://github.com/tinkoff-ai/etna/pull/1210))
 - Add default `params_to_tune` for `HoltWintersModel`, `HoltModel` and `SimpleExpSmoothingModel` ([#1209](https://github.com/tinkoff-ai/etna/pull/1209))
 - Add default `params_to_tune` for `RNNModel` and `MLPModel` ([#1218](https://github.com/tinkoff-ai/etna/pull/1218))
+- Add default `params_to_tune` for `DateFlagsTransform`, `TimeFlagsTransform`, `SpecialDaysTransform` and `FourierTransform` ([#1228](https://github.com/tinkoff-ai/etna/pull/1228))
 ### Fixed
 - Fix bug in `GaleShapleyFeatureSelectionTransform` with wrong number of remaining features ([#1110](https://github.com/tinkoff-ai/etna/pull/1110))
 - `ProphetModel` fails with additional seasonality set ([#1157](https://github.com/tinkoff-ai/etna/pull/1157))

etna/transforms/timestamp/date_flags.py

@@ -1,15 +1,21 @@
 from copy import deepcopy
 from math import ceil
+from typing import Dict
 from typing import List
 from typing import Optional
 from typing import Sequence
 
 import numpy as np
 import pandas as pd
 
+from etna import SETTINGS
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversibleTransform
 
+if SETTINGS.auto_required:
+    from optuna.distributions import BaseDistribution
+    from optuna.distributions import CategoricalDistribution
+
 
 class DateFlagsTransform(IrreversibleTransform, FutureMixin):
     """DateFlagsTransform is a class that implements extraction of the main date-based features from datetime column.
@@ -345,5 +351,27 @@ def _get_weekends(timestamp_series: pd.Series) -> np.ndarray:
         weekend_days = (5, 6)
         return timestamp_series.apply(lambda x: x.weekday() in weekend_days).values
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        There are no restrictions on all ``False`` values for the flags.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return {
+            "day_number_in_week": CategoricalDistribution([False, True]),
+            "day_number_in_month": CategoricalDistribution([False, True]),
+            "day_number_in_year": CategoricalDistribution([False, True]),
+            "week_number_in_month": CategoricalDistribution([False, True]),
+            "week_number_in_year": CategoricalDistribution([False, True]),
+            "month_number_in_year": CategoricalDistribution([False, True]),
+            "season_number": CategoricalDistribution([False, True]),
+            "year_number": CategoricalDistribution([False, True]),
+            "is_weekend": CategoricalDistribution([False, True]),
+        }
+
 
 __all__ = ["DateFlagsTransform"]

etna/transforms/timestamp/fourier.py

@@ -1,14 +1,20 @@
 import math
+from typing import Dict
 from typing import List
 from typing import Optional
 from typing import Sequence
 
 import numpy as np
 import pandas as pd
 
+from etna import SETTINGS
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversibleTransform
 
+if SETTINGS.auto_required:
+    from optuna.distributions import BaseDistribution
+    from optuna.distributions import IntLogUniformDistribution
+
 
 class FourierTransform(IrreversibleTransform, FutureMixin):
     """Adds fourier features to the dataset.
@@ -72,19 +78,22 @@ def __init__(
         if period < 2:
             raise ValueError("Period should be at least 2")
         self.period = period
-        self.mods: Sequence[int]
+
+        self.order = order
+        self.mods = mods
+        self._mods: Sequence[int]
 
         if order is not None and mods is None:
             if order < 1 or order > math.ceil(period / 2):
                 raise ValueError("Order should be within [1, ceil(period/2)] range")
-            self.mods = [mod for mod in range(1, 2 * order + 1) if mod < period]
+            self._mods = [mod for mod in range(1, 2 * order + 1) if mod < period]
         elif mods is not None and order is None:
             if min(mods) < 1 or max(mods) >= period:
                 raise ValueError("Every mod should be within [1, int(period)) range")
-            self.mods = mods
+            self._mods = mods
         else:
             raise ValueError("There should be exactly one option set: order or mods")
-        self.order = None
+
         self.out_column = out_column
         super().__init__(required_features=["target"])
 
@@ -96,7 +105,7 @@ def _get_column_name(self, mod: int) -> str:
 
     def get_regressors_info(self) -> List[str]:
         """Return the list with regressors created by the transform."""
-        output_columns = [self._get_column_name(mod=mod) for mod in self.mods]
+        output_columns = [self._get_column_name(mod=mod) for mod in self._mods]
         return output_columns
 
     def _fit(self, df: pd.DataFrame) -> "FourierTransform":
@@ -143,10 +152,28 @@ def _transform(self, df: pd.DataFrame) -> pd.DataFrame:
         features = pd.DataFrame(index=df.index)
         elapsed = np.arange(features.shape[0]) / self.period
 
-        for mod in self.mods:
+        for mod in self._mods:
             order = (mod + 1) // 2
             is_cos = mod % 2 == 0
 
             features[self._get_column_name(mod)] = np.sin(2 * np.pi * order * elapsed + np.pi / 2 * is_cos)
 
         return self._construct_answer(df, features)
+
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        * If the ``mods`` parameter is set, then the grid is empty.
+
+        * If the ``order`` parameter is set, ``period`` parameter isn't tuned. It is expected to be set by the user.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        if self.mods is not None:
+            return {}
+
+        max_value = math.ceil(self.period / 2)
+        return {"order": IntLogUniformDistribution(low=1, high=max_value)}

etna/transforms/timestamp/special_days.py

@@ -7,10 +7,15 @@
 
 import pandas as pd
 
+from etna import SETTINGS
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversiblePerSegmentWrapper
 from etna.transforms.base import OneSegmentTransform
 
+if SETTINGS.auto_required:
+    from optuna.distributions import BaseDistribution
+    from optuna.distributions import CategoricalDistribution
+
 
 def calc_day_number_in_week(datetime_day: datetime.datetime) -> int:
     return datetime_day.weekday()
@@ -216,5 +221,20 @@ def get_regressors_info(self) -> List[str]:
             output_columns.append("anomaly_monthdays")
         return output_columns
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        There are no restrictions on all ``False`` values for the flags.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return {
+            "find_special_weekday": CategoricalDistribution([False, True]),
+            "find_special_month_day": CategoricalDistribution([False, True]),
+        }
+
 
 __all__ = ["SpecialDaysTransform"]

etna/transforms/timestamp/time_flags.py

@@ -1,13 +1,19 @@
 from copy import deepcopy
+from typing import Dict
 from typing import List
 from typing import Optional
 
 import numpy as np
 import pandas as pd
 
+from etna import SETTINGS
 from etna.transforms.base import FutureMixin
 from etna.transforms.base import IrreversibleTransform
 
+if SETTINGS.auto_required:
+    from optuna.distributions import BaseDistribution
+    from optuna.distributions import CategoricalDistribution
+
 
 class TimeFlagsTransform(IrreversibleTransform, FutureMixin):
     """TimeFlagsTransform is a class that implements extraction of the main time-based features from datetime column."""
@@ -202,5 +208,24 @@ def _get_period_in_day(timestamp_series: pd.Series, period_in_hours: int = 12) -
         """
         return timestamp_series.apply(lambda x: x.hour // period_in_hours).values
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        There are no restrictions on all ``False`` values for the flags.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return {
+            "minute_in_hour_number": CategoricalDistribution([False, True]),
+            "fifteen_minutes_in_hour_number": CategoricalDistribution([False, True]),
+            "hour_number": CategoricalDistribution([False, True]),
+            "half_hour_number": CategoricalDistribution([False, True]),
+            "half_day_number": CategoricalDistribution([False, True]),
+            "one_third_day_number": CategoricalDistribution([False, True]),
+        }
+
 
 __all__ = ["TimeFlagsTransform"]

tests/test_pipeline/test_autoregressive_pipeline.py

@@ -7,6 +7,7 @@
 import numpy as np
 import pandas as pd
 import pytest
+from optuna.distributions import CategoricalDistribution
 from optuna.distributions import IntUniformDistribution
 from optuna.distributions import LogUniformDistribution
 
@@ -394,6 +395,15 @@ def test_predict_return_components(example_tsds, model_fixture, request):
                 "model.depth": IntUniformDistribution(low=1, high=11, step=1),
                 "model.l2_leaf_reg": LogUniformDistribution(low=0.1, high=200.0),
                 "model.random_strength": LogUniformDistribution(low=1e-05, high=10.0),
+                "transforms.0.day_number_in_week": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.month_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.season_number": CategoricalDistribution([False, True]),
+                "transforms.0.year_number": CategoricalDistribution([False, True]),
+                "transforms.0.is_weekend": CategoricalDistribution([False, True]),
             },
         ),
     ],

tests/test_pipeline/test_hierarchical_pipeline.py

@@ -5,6 +5,9 @@
 import numpy as np
 import pandas as pd
 import pytest
+from optuna.distributions import CategoricalDistribution
+from optuna.distributions import IntUniformDistribution
+from optuna.distributions import LogUniformDistribution
 
 from etna.datasets.utils import match_target_quantiles
 from etna.metrics import MAE
@@ -441,3 +444,45 @@ def test_raw_forecast_with_return_components(product_level_constant_hierarchical
     )
     pipeline.fit(product_level_constant_hierarchical_ts)
     pipeline.raw_forecast(ts=product_level_constant_hierarchical_ts, return_components=True)
+
+
+@pytest.mark.parametrize(
+    "reconciliator",
+    (
+        TopDownReconciliator(target_level="product", source_level="market", period=1, method="AHP"),
+        TopDownReconciliator(target_level="product", source_level="market", period=1, method="PHA"),
+        BottomUpReconciliator(target_level="market", source_level="product"),
+        BottomUpReconciliator(target_level="total", source_level="market"),
+    ),
+)
+@pytest.mark.parametrize(
+    "model, transforms, expected_params_to_tune",
+    [
+        (
+            CatBoostMultiSegmentModel(iterations=100),
+            [DateFlagsTransform(), LagTransform(in_column="target", lags=list(range(3, 10)))],
+            {
+                "model.learning_rate": LogUniformDistribution(low=1e-4, high=0.5),
+                "model.depth": IntUniformDistribution(low=1, high=11, step=1),
+                "model.l2_leaf_reg": LogUniformDistribution(low=0.1, high=200.0),
+                "model.random_strength": LogUniformDistribution(low=1e-05, high=10.0),
+                "transforms.0.day_number_in_week": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.month_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.season_number": CategoricalDistribution([False, True]),
+                "transforms.0.year_number": CategoricalDistribution([False, True]),
+                "transforms.0.is_weekend": CategoricalDistribution([False, True]),
+            },
+        ),
+    ],
+)
+def test_params_to_tune(reconciliator, model, transforms, expected_params_to_tune):
+    horizon = 1
+    pipeline = HierarchicalPipeline(reconciliator=reconciliator, model=model, transforms=transforms, horizon=horizon)
+
+    obtained_params_to_tune = pipeline.params_to_tune()
+
+    assert obtained_params_to_tune == expected_params_to_tune

tests/test_pipeline/test_pipeline.py

@@ -7,6 +7,7 @@
 import numpy as np
 import pandas as pd
 import pytest
+from optuna.distributions import CategoricalDistribution
 from optuna.distributions import IntUniformDistribution
 from optuna.distributions import LogUniformDistribution
 
@@ -1221,6 +1222,15 @@ def test_predict_return_components(
                 "model.depth": IntUniformDistribution(low=1, high=11, step=1),
                 "model.l2_leaf_reg": LogUniformDistribution(low=0.1, high=200.0),
                 "model.random_strength": LogUniformDistribution(low=1e-05, high=10.0),
+                "transforms.0.day_number_in_week": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.day_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_month": CategoricalDistribution([False, True]),
+                "transforms.0.week_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.month_number_in_year": CategoricalDistribution([False, True]),
+                "transforms.0.season_number": CategoricalDistribution([False, True]),
+                "transforms.0.year_number": CategoricalDistribution([False, True]),
+                "transforms.0.is_weekend": CategoricalDistribution([False, True]),
             },
         ),
     ],

tests/test_transforms/test_timestamp/test_dateflags_transform.py

@@ -11,6 +11,7 @@
 
 from etna.datasets import TSDataset
 from etna.transforms.timestamp import DateFlagsTransform
+from tests.test_transforms.utils import assert_sampling_is_valid
 from tests.test_transforms.utils import assert_transformation_equals_loaded_original
 
 WEEKEND_DAYS = (5, 6)
@@ -289,3 +290,10 @@ def test_save_load(train_ts):
     ts = train_ts
     transform = DateFlagsTransform()
     assert_transformation_equals_loaded_original(transform=transform, ts=ts)
+
+
+def test_params_to_tune(train_ts):
+    transform = DateFlagsTransform()
+    ts = train_ts
+    assert len(transform.params_to_tune()) > 0
+    assert_sampling_is_valid(transform=transform, ts=ts)

tests/test_transforms/test_timestamp/test_fourier_transform.py

@@ -8,6 +8,7 @@
 from etna.metrics import R2
 from etna.models import LinearPerSegmentModel
 from etna.transforms.timestamp import FourierTransform
+from tests.test_transforms.utils import assert_sampling_is_valid
 from tests.test_transforms.utils import assert_transformation_equals_loaded_original
 
 
@@ -49,15 +50,15 @@ def ts_trend_seasonal(random_seed) -> TSDataset:
     return TSDataset(TSDataset.to_dataset(classic_df), freq="D")
 
 
-@pytest.mark.parametrize("order, mods, repr_mods", [(None, [1, 2, 3, 4], [1, 2, 3, 4]), (2, None, [1, 2, 3, 4])])
-def test_repr(order, mods, repr_mods):
+@pytest.mark.parametrize("order, mods", [(None, [1, 2, 3, 4]), (2, None)])
+def test_repr(order, mods):
     transform = FourierTransform(
         period=10,
         order=order,
         mods=mods,
     )
     transform_repr = transform.__repr__()
-    true_repr = f"FourierTransform(period = 10, order = None, mods = {repr_mods}, out_column = None, )"
+    true_repr = f"FourierTransform(period = 10, order = {order}, mods = {mods}, out_column = None, )"
     assert transform_repr == true_repr
 
 
@@ -166,3 +167,19 @@ def test_forecast(ts_trend_seasonal):
 def test_save_load(ts_trend_seasonal):
     transform = FourierTransform(period=7, order=3)
     assert_transformation_equals_loaded_original(transform=transform, ts=ts_trend_seasonal)
+
+
+@pytest.mark.parametrize(
+    "transform, expected_length",
+    [
+        (FourierTransform(period=7, order=1), 1),
+        (FourierTransform(period=7, mods=[1]), 0),
+        (FourierTransform(period=7, mods=[1, 4]), 0),
+        (FourierTransform(period=30.4, order=1), 1),
+        (FourierTransform(period=365.25, order=1), 1),
+    ],
+)
+def test_params_to_tune(transform, expected_length, ts_trend_seasonal):
+    ts = ts_trend_seasonal
+    assert len(transform.params_to_tune()) == expected_length
+    assert_sampling_is_valid(transform=transform, ts=ts)