sb-ai-lab / automlwhitebox Goto Github PK

All feature history from Selector (Metrics, Correlation and L1) are rewriting by last model refit.

Example:

from sklearn.datasets import load_breast_cancer
from autowoe import AutoWoE

X, y = load_breast_cancer(return_X_y=True, as_frame=True)

test_aw = AutoWoE(n_jobs=1, debug=True)
test_aw.fit(pd.concat([X, y], axis=1), 'target')

If we look at test_aw.feature_history we see a lot of 'Pruned during regression refit' reason and nothing is about selectors. But we know exactly that selector's reason exists. How did I check this:

history = {k: None if v == 'Pruned during regression refit' else v for k, v in test_aw.feature_history.items()}

selector = Selector(
    interpreted_model=test_aw.params["interpreted_model"],
    task=test_aw.params["task"],
    train=test_aw.train_df,
    target=test_aw.target,
    features_type=test_aw.private_features_type,
    n_jobs=test_aw.params["n_jobs"],
    cv_split=test_aw._cv_split,
    features_mark_values=None,
)

best_features, _sel_result = selector(
    history,
    pearson_th=test_aw.params["pearson_th"],
    metric_th=test_aw.params["metric_th"],
    vif_th=test_aw.params["vif_th"],
    l1_grid_size=test_aw.params["l1_grid_size"],
    l1_exp_scale=test_aw.params['l1_exp_scale'],
    metric_tol=test_aw.params["metric_tol"],
)

And if we look at history now we can see multiply different drop reason

My suggestion is to change third argument for last feature_changing() in AutoWoE.fit() from self._private_features_type to best_features, because now feature_changing() thinks that all input features for selectors is features_before for last refit.

This check results in an error:
"module 'numpy' has no attribute 'str'.
np.str was a deprecated alias for the builtin str. To avoid this error in existing code, use str by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use np.str_ here.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations"