How do I get best 'score' coefs instead of best 'estimator'?

Question

I have created my own estimator that wraps a sklearn regressor.

GSV = GridSearchCV(pipe,
                       param_grid = {'step2':[1,2,3]},
                       cv=splits,
                       verbose=1
                       )

GSV.fit(X, y)

I usually look at :

GSV.best_estimator_.named_steps['step2'].coef_

To get best coefs.

However, I may score an estimator result to zero in the 'score' override. This appears to be applied after best.estimator has been set - how can I get coefs for my best score?