clintk.cat2vec.feature_selection module

selects parameters with L1 logistic regression

class clintk.cat2vec.feature_selection.LassoSelector(lasso_coefs, feature_col, coef_col, n_features=64)

Bases: sklearn.base.BaseEstimator

This class is made to be used after cat2vec.lasso_gridsearch since it selects the features from a dataframe that have the most weighted coefficients (according to a L1-penalized linear model)

It inherits from sklearn.base.BaseEstimator to allow gridsearching the best n_features using a pipeline and a basline classifier

Parameters:

• n_features (int) – number of top features to keep
• lasso_coefs (pd.DataFrame) – each row is the name of a category and its coef weight in LASSO model
• feature_col (str) – name of the feature col (ie name of the categorical variable)
• coef_col (str) – name of the column of the LASSO coefficients in lasso_coefs dataframe

Examples

>>> dico = {'coef': [0, 4.5, 1.2, 0.3],                 'colnames': ['feat1', 'feat2', 'feat3', 'feat4']}
>>> df = pd.DataFrame(dico)
keeps only feat2 and feat3
>>> selector = LassoSelector(2).fit(df['colnames'], df['coef'])
>>> X = [[0, 0, 1, 0], [1, 1, 0, 0], [0, 1, 0, 0]]
>>> selector.transform(X)
[[0, 1], [1, 0], [1, 0]]

fit(X, y)

transform(X)

Parameters:X (pd.DataFrame) – contains only features Returns:contains the best n_features Return type:ndarray