Add params_to_tune for linear models (#1204)

CHANGELOG.md

@@ -45,6 +45,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Remove version python-3.7 from `pyproject.toml`, update lock ([#1183](https://github.com/tinkoff-ai/etna/pull/1183))
 - Add default `params_to_tune` for catboost models ([#1185](https://github.com/tinkoff-ai/etna/pull/1185))
 - Add default `params_to_tune` for `ProphetModel` ([#1203](https://github.com/tinkoff-ai/etna/pull/1203))
+- Add default `params_to_tune` for linear models ([#1204](https://github.com/tinkoff-ai/etna/pull/1204))
 ### 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/models/linear.py

@@ -1,13 +1,33 @@
+from typing import Dict
+
 import pandas as pd
 from sklearn.linear_model import ElasticNet
 from sklearn.linear_model import LinearRegression
 
+from etna import SETTINGS
 from etna.models.base import NonPredictionIntervalContextIgnorantAbstractModel
 from etna.models.mixins import MultiSegmentModelMixin
 from etna.models.mixins import NonPredictionIntervalContextIgnorantModelMixin
 from etna.models.mixins import PerSegmentModelMixin
 from etna.models.sklearn import _SklearnAdapter
 
+if SETTINGS.auto_required:
+    from optuna.distributions import BaseDistribution
+    from optuna.distributions import CategoricalDistribution
+    from optuna.distributions import LogUniformDistribution
+    from optuna.distributions import UniformDistribution
+
+
+LINEAR_GRID: Dict[str, "BaseDistribution"] = {
+    "fit_intercept": CategoricalDistribution([False, True]),
+}
+
+ELASTIC_GRID: Dict[str, "BaseDistribution"] = {
+    "fit_intercept": CategoricalDistribution([False, True]),
+    "l1_ratio": UniformDistribution(0, 1),
+    "alpha": LogUniformDistribution(low=1e-5, high=1e3),
+}
+
 
 class _LinearAdapter(_SklearnAdapter):
     def predict_components(self, df: pd.DataFrame) -> pd.DataFrame:
@@ -64,6 +84,16 @@ def __init__(self, fit_intercept: bool = True, **kwargs):
             base_model=_LinearAdapter(regressor=LinearRegression(fit_intercept=self.fit_intercept, **self.kwargs))
         )
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return LINEAR_GRID
+
 
 class ElasticPerSegmentModel(
     PerSegmentModelMixin,
@@ -117,6 +147,16 @@ def __init__(self, alpha: float = 1.0, l1_ratio: float = 0.5, fit_intercept: boo
             )
         )
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return ELASTIC_GRID
+
 
 class LinearMultiSegmentModel(
     MultiSegmentModelMixin,
@@ -147,6 +187,16 @@ def __init__(self, fit_intercept: bool = True, **kwargs):
             base_model=_LinearAdapter(regressor=LinearRegression(fit_intercept=self.fit_intercept, **self.kwargs))
         )
 
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return LINEAR_GRID
+
 
 class ElasticMultiSegmentModel(
     MultiSegmentModelMixin,
@@ -199,3 +249,13 @@ def __init__(self, alpha: float = 1.0, l1_ratio: float = 0.5, fit_intercept: boo
                 )
             )
         )
+
+    def params_to_tune(self) -> Dict[str, "BaseDistribution"]:
+        """Get default grid for tuning hyperparameters.
+
+        Returns
+        -------
+        :
+            Grid to tune.
+        """
+        return ELASTIC_GRID

tests/test_models/test_linear_model.py

@@ -4,6 +4,7 @@
 import numpy as np
 import pandas as pd
 import pytest
+from optuna.samplers import RandomSampler
 from sklearn.linear_model import ElasticNet
 from sklearn.linear_model import LinearRegression
 
@@ -325,3 +326,17 @@ def test_linear_adapter_predict_components_sum_up_to_target(df_with_regressors,
     target = adapter.predict(df)
     target_components = adapter.predict_components(df)
     np.testing.assert_array_almost_equal(target, target_components.sum(axis=1), decimal=10)
+
+
+@pytest.mark.parametrize(
+    "model", [LinearPerSegmentModel(), LinearMultiSegmentModel(), ElasticPerSegmentModel(), ElasticMultiSegmentModel()]
+)
+def test_params_to_tune(model):
+    grid = model.params_to_tune()
+    # we need sampler to get a value from distribution
+    sampler = RandomSampler()
+
+    assert len(grid) > 0
+    for name, distribution in grid.items():
+        value = sampler.sample_independent(study=None, trial=None, param_name=name, param_distribution=distribution)
+        _ = model.set_params(**{name: value})