Feature selection

CHANGELOG.md

@@ -10,6 +10,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `DeseasonalityTransform` ([#1307](https://github.com/tinkoff-ai/etna/pull/1307))
 - 
 - Add extension with models from `statsforecast`: `StatsForecastARIMAModel`, `StatsForecastAutoARIMAModel`, `StatsForecastAutoCESModel`, `StatsForecastAutoETSModel`, `StatsForecastAutoThetaModel` ([#1295](https://github.com/tinkoff-ai/etna/pull/1297))
+- Notebook `feature_selection` ([#875](https://github.com/tinkoff-ai/etna/pull/875))
 - 
 - 
 

README.md

@@ -175,15 +175,19 @@ We have also prepared a set of tutorials for an easy introduction:
 | [Get started](https://github.com/tinkoff-ai/etna/tree/master/examples/get_started.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/get_started.ipynb) |
 | [Backtest](https://github.com/tinkoff-ai/etna/tree/master/examples/backtest.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/backtest.ipynb) |
 | [EDA](https://github.com/tinkoff-ai/etna/tree/master/examples/EDA.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/EDA.ipynb) |
-| [Outliers](https://github.com/tinkoff-ai/etna/tree/master/examples/outliers.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/outliers.ipynb) |
-| [Clustering](https://github.com/tinkoff-ai/etna/tree/master/examples/clustering.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/clustering.ipynb) |
+| [Regressors and exogenous data](https://github.com/tinkoff-ai/etna/tree/master/examples/exogenous_data.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/exogenous_data.ipynb) |
+| [Custom model and transform](https://github.com/tinkoff-ai/etna/tree/master/examples/custom_transform_and_model.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/custom_transform_and_model.ipynb) |
 | [Deep learning models](https://github.com/tinkoff-ai/etna/tree/master/examples/NN_examples.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/NN_examples.ipynb) |
 | [Ensembles](https://github.com/tinkoff-ai/etna/tree/master/examples/ensembles.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/ensembles.ipynb) |
-| [Custom Transform and Model](https://github.com/tinkoff-ai/etna/tree/master/examples/custom_transform_and_model.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/custom_transform_and_model.ipynb) |
-| [Exogenous data](https://github.com/tinkoff-ai/etna/tree/master/examples/exogenous_data.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/exogenous_data.ipynb) |
-| [Forecasting strategies](https://github.com/tinkoff-ai/etna/blob/master/examples/forecasting_strategies.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/forecasting_strategies.ipynb) |
-| [Classification](https://github.com/tinkoff-ai/etna/blob/master/examples/classification.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/classification.ipynb) |
+| [Outliers](https://github.com/tinkoff-ai/etna/tree/master/examples/outliers.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/outliers.ipynb) |
+| [Forecasting strategies](https://github.com/tinkoff-ai/etna/tree/master/examples/forecasting_strategies.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/forecasting_strategies.ipynb) |
+| [Forecast interpretation](https://github.com/tinkoff-ai/etna/tree/master/examples/forecast_interpretation.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/forecast_interpretation.ipynb) |
+| [Clustering](https://github.com/tinkoff-ai/etna/tree/master/examples/clustering.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/clustering.ipynb) |
+| [AutoML](https://github.com/tinkoff-ai/etna/tree/master/examples/automl.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/automl.ipynb) |
+| [Inference: using saved pipeline on a new data](https://github.com/tinkoff-ai/etna/tree/master/examples/inference.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/inference.ipynb) |
 | [Hierarchical time series](https://github.com/tinkoff-ai/etna/blob/master/examples/hierarchical_pipeline.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/hierarchical_pipeline.ipynb) |
+| [Classification](https://github.com/tinkoff-ai/etna/blob/master/examples/classification.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/classification.ipynb) |
+| [Feature selection](https://github.com/tinkoff-ai/etna/blob/master/examples/feature_selection.ipynb) | [![Binder](https://mybinder.org/badge_logo.svg)](https://mybinder.org/v2/gh/tinkoff-ai/etna/master?filepath=examples/feature_selection.ipynb) |
 
 ## Documentation
 

etna/analysis/feature_relevance/relevance_table.py

@@ -36,6 +36,7 @@ def _prepare_df(df: pd.DataFrame, df_exog: pd.DataFrame, segment: str, regressor
 
 def get_statistics_relevance_table(df: pd.DataFrame, df_exog: pd.DataFrame) -> pd.DataFrame:
     """Calculate relevance table with p-values from tsfresh.
+
     Parameters
     ----------
     df:

etna/transforms/feature_selection/gale_shapley.py

@@ -87,7 +87,7 @@ def get_next_candidate(self) -> Optional[str]:
 
         Returns
         -------
-        name: str
+        name:
             name of feature
         """
         if self.last_candidate is None:
@@ -113,7 +113,7 @@ def check_segment(self, segment: str) -> bool:
 
         Returns
         -------
-        is_better: bool
+        is_better:
             returns True if given segment is a better candidate than current match.
         """
         if self.tmp_match is None or self.tmp_match_rank is None:
@@ -178,7 +178,7 @@ def _gale_shapley_iteration(self, available_segments: List[SegmentGaleShapley])
 
         Returns
         -------
-        success: bool
+        success:
             True if there is at least one match attempt at the iteration
 
         Notes
@@ -212,7 +212,7 @@ def __call__(self) -> Dict[str, str]:
 
         Returns
         -------
-        matching: Dict[str, str]
+        matching:
             matching dict of segment x feature
         """
         success_run = True
@@ -224,13 +224,23 @@ def __call__(self) -> Dict[str, str]:
 
 
 class GaleShapleyFeatureSelectionTransform(BaseFeatureSelectionTransform):
-    """GaleShapleyFeatureSelectionTransform provides feature filtering with Gale-Shapley matching algo according to relevance table.
-
+    """Transform that provides feature filtering by Gale-Shapley matching algorithm according to the relevance table.
 
     Notes
     -----
     Transform works with any type of features, however most of the models works only with regressors.
     Therefore, it is recommended to pass the regressors into the feature selection transforms.
+
+    As input, we have a table of relevances with size :math:`N\_{f} \times N\_{s}` where :math:`N\_{f}` -- number of features,
+    :math:`N\_{s}` -- number of segments.
+    Procedure of filtering features consist of :math:`\lceil \frac{k}{N\_{s}} \rceil` iterations.
+    Algorithm of each iteration:
+
+    - build a matching between segments and features by `Gale–Shapley algorithm <https://en.wikipedia.org/wiki/Gale%E2%80%93Shapley_algorithm>`_
+    according to the relevance table, during the matching segments send proposals to features;
+    - select features to add by taking matched feature for each segment;
+    - add selected features to accumulated list of selected features taking into account that this list shouldn't exceed the size of ``top_k``;
+    - remove added features from future consideration.
     """
 
     def __init__(
@@ -290,7 +300,8 @@ def _compute_gale_shapley_steps_number(top_k: int, n_segments: int, n_features:
             return 1
         if top_k < n_segments:
             warnings.warn(
-                f"Given top_k={top_k} is less than n_segments. Algo will filter data without Gale-Shapley run."
+                f"Given top_k={top_k} is less than n_segments={n_segments}. "
+                f"Algo will filter data without Gale-Shapley run."
             )
             return 1
         return ceil(top_k / n_segments)
@@ -309,7 +320,7 @@ def _gale_shapley_iteration(
 
         Returns
         -------
-        matching dict: Dict[str, str]
+        matching dict:
             dict of segment x feature
         """
         gssegments = [

examples/README.md

@@ -15,17 +15,17 @@ We have prepared a set of tutorials for an easy introduction:
 
 #### 03. [EDA](https://github.com/tinkoff-ai/etna/tree/master/examples/EDA.ipynb) 
 - Visualization
-    - Plot
-    - Partial autocorrelation
-    - Cross-correlation
-    - Correlation heatmap
-    - Distribution
+  - Plot
+  - Partial autocorrelation
+  - Cross-correlation
+  - Correlation heatmap
+  - Distribution
 - Outliers
-     - Median method
-     - Density method
+  - Median method
+  - Density method
 - Change Points
-     - Change points plot
-     - Interactive change points plot
+  - Change points plot
+  - Interactive change points plot
 
 #### 04. [Regressors and exogenous data](https://github.com/tinkoff-ai/etna/tree/master/examples/exogenous_data.ipynb)
 - What is regressor? 
@@ -35,7 +35,7 @@ We have prepared a set of tutorials for an easy introduction:
   - EDA
 - Forecast with regressors
 
-#### 05. [Custom model and transform](https://github.com/tinkoff-ai/etna/tree/master/examples/exogenous_data.ipynb)
+#### 05. [Custom model and transform](https://github.com/tinkoff-ai/etna/tree/master/examples/custom_transform_and_model.ipynb)
 - What is Transform and how it works 
 - Custom Transform 
   - Per-segment Custom Transform 
@@ -56,10 +56,10 @@ We have prepared a set of tutorials for an easy introduction:
 
 #### 08. [Outliers](https://github.com/tinkoff-ai/etna/tree/master/examples/outliers.ipynb) 
 - Point outliers
-    - Median method
-    - Density method
-    - Prediction interval method
-    - Histogram method
+  - Median method
+  - Density method
+  - Prediction interval method
+  - Histogram method
 - Sequence outliers
 - Interactive visualization
 - Outliers imputation
@@ -96,11 +96,11 @@ We have prepared a set of tutorials for an easy introduction:
 
 #### 13. [AutoML notebook](https://github.com/tinkoff-ai/etna/tree/master/examples/automl.ipynb)
 - Hyperparameters tuning
-    - How `Tune` works
-    - Example
+  - How `Tune` works
+  - Example
 - General AutoML
-    - How `Auto` works
-    - Example
+  - How `Auto` works
+  - Example
 
 #### 14. Hyperparameter search
 - [Optuna](https://github.com/tinkoff-ai/etna/tree/master/examples/optuna)
@@ -115,3 +115,21 @@ We have prepared a set of tutorials for an easy introduction:
 - Hierarchical structure
 - Reconciliation methods
 - Exogenous variables for hierarchical forecasts
+
+#### 17. [Classification](https://github.com/tinkoff-ai/etna/tree/master/examples/classification.ipynb)
+- Classification 
+  - Load Dataset
+  - Feature extraction 
+  - Cross validation 
+- Predictability analysis
+  - Load Dataset 
+  - Load pretrained analyzer 
+  - Analyze segments predictability
+
+#### 18. [Feature selection](https://github.com/tinkoff-ai/etna/tree/master/examples/feature_selection.ipynb)
+- Loading Dataset
+- Feature selection methods
+  - Intro to feature selection
+  - `TreeFeatureSelectionTransform`
+  - `GaleShapleyFeatureSelectionTransform`
+  - `MRMRFeatureSelectionTransform`