fix: Create Visualizer from a copy of the regressor to prevent updates to original regressor

[SiddhantSadangi]

create_feature_importance_chart() would instantiate FeatureImportances() on the original regressor and then refit the instantiated object. In some cases (as shown in #29), this can update the coefficients of the original model.

This change instantiates FeatureImportances() on a deepcopy of the original regressor, preventing any changes to the original.

[AleksanderWWW]

@SiddhantSadangi could you add a test case that would confirm that the regressor is indeed not changed after running the create_feature_importance_chart?

[SiddhantSadangi]

@AleksanderWWW - clubbed it with test_regressor_summary. Lemme know if you want me to create a separate test instead

[AleksanderWWW]

@AleksanderWWW - clubbed it with test_regressor_summary. Lemme know if you want me to create a separate test instead

Up to you, if you feel that it would be worth it then go ahead, but if it covers what needs to be covered already, then leave it as is

Sorry, I wanted to reply and accidentally edited your comment XD

[AleksanderWWW]

Also, check out this https://github.com/scikit-learn/scikit-learn/blob/70fdc843a4b8182d97a3508c1a426acc5e87e980/sklearn/base.py#L40

perhaps we could delegate the copying process to sklearn

[SiddhantSadangi]

sklearn.base.clone() returns an unfitted estimator. It would require us to pass is_fitted=False when instantiating the visualizer. This will require the estimator to be fit as well when the visualizer is fit, increasing runtime.