SHFS 0.1.5

Feature election group of classes calculate the importance of features based on the Shap library for the classification and regression problem Only works with randomforest models for efficiency or gradient boosting models. DFwrapper - remove multicollinearity and outliers

Library consist of two groups of Classes - Feature selectors and DFwrapper to have a deal with outliers and correlation

1. Feature selection group

The FeatureSelection calculates the importance of features based on the Shap library for a classification problem. Only works with trees for better efficiency or models based on gradient boosting. It is a priority to use such models as:

Catboost - does not require handling of NaN and categories. works with sklearn

NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

        !pip install shap 
        !pip install phik

https://pypi.org/project/SHFS/

        FeatureSelectionClf - for classification
        FeatureSelectionRegression - for regression
        FeatureSelectionUniversal - for both classification and regression tasks

Quick start: Collab and Tutorial

Parametrs.

---

`estimator` :   
    Supervised learning with the fit method will allow you to retrieve and select indices.
    the most important features.
n_features_to_select: int, default = None.
    The number of features to select, the default is None.
columns: List, default = None.
    The list of attributes of the initial set, the default is None.

Methods

---

fit - trains and identifies the most important features
tranform - changes the original set and returns the selected attributes
get_index - Returns the selected indexes attributes

only for FeatureSelectionClf and FeatureSelectionRegression:

plot_values - plotting shap values
_estimator_type - @property method 
get_feature_importance - Returns DataFrame FI

Note

---

Nan / Inf are allowed in case they are accepted by the fit method model Example use for classification

---

cols = list(X_train.columns)
cat_features = list(X_train_cat.select_dtypes(include=['object', 'category']).columns)
num_features = list(X_train_cat.select_dtypes(exclude=['object', 'category']).columns)
estimator = CatBoostClassifier(**params_cat)
selector = FeatureSelectionClf(estimator, n_features_to_select=3, columns=cols) 
preprocessor = ColumnTransformer (
    transformers = [

        ('std_scaler' , StandardScaler() , num_features) ,
        ('cat' , OrdinalEncoder() , cat_features),
        
        ]
)

pipe = Pipeline(steps=
    
    [ 
      ('preprocessor', preprocessor),
      ('selector', selector)

    ]
)
   X_train_prep = pipe.fit_transform(X_train)

Example without Pipeline

   cols = list(X_train.columns)
   estimator = CatBoostClassifier(**params_cat)
   selector = FeatureSelectionClf(estimator, n_features_to_select=3, columns=cols)
   X = selector.fit(X_train_prep, y_train)

2. DFwrapper

DFwrapper - remove multicollinearity and outliers from Pandas DataFrame

Usage example
----------
1. Collinearity

cleaner = DFwrapper()
new_df = cleaner.wrap_corr(df)

2. Outliers. Rough cleaning

cleaner = DFwrapper(low=.05, high=.95)
cleaned = cleaner.quantile_cleaner(df, cols_to_clean)

2. Outliers. Finer cleaning

cleaner = DFwrapper(koeff=1.5)
cleaned = cleaner.frame_irq(df, cols_to_clean)