def fit(self, X, y=None):
X_norm = MaxAbsScaler().fit_transform(X)
if issparse(X):
if type(X) != csc_matrix:
X = X.tocsc()
X_norm = X_norm.tocsc()
print('Running Cor')
cor_support = cor_selector(X, y, self.feature_names, self.num_features)
print('Running Chi2')
chi_support = chi2_selector(X_norm, y, self.num_features)
print('Running RFE')
rfe_support = rfe_selector(X_norm, y, self.num_features,
self.random_state)
print('Running LR')
embeded_lr_support = embeded_lr_selector(X_norm, y, self.num_features,
self.random_state)
print('Running RF')
embeded_rf_support = embeded_rf_selector(
X,
y,
self.num_features,
n_jobs=self.n_jobs,
random_state=self.random_state)
print('Running XG')
embeded_xgb_support = embeded_xgb_selector(
X,
y,
self.num_features,
n_jobs=self.n_jobs,
random_state=self.random_state)
feature_selection_df = pd.DataFrame({
'feature': self.feature_names,
'pearson': cor_support,
'chi_2': chi_support,
'rfe': rfe_support,
'logistics': embeded_lr_support,
'random_forest': embeded_rf_support,
'xgboost': embeded_xgb_support
})
feature_selection_df['total'] = np.sum(feature_selection_df, axis=1)
self.feature_selection_df_ = feature_selection_df
keep_features = feature_selection_df.query('total >= {}'.format(
self.min_selections))['feature'].tolist()
# Keep the features that we always want (e.g. domain expertise)
if self.always_keep is not None:
keep_features.extend(self.always_keep)
self.keep_features_ = [
f for f in self.feature_names if f in keep_features
]
return self
