Python: Ridge regression - ''Ridge' object has no attribute 'coef_' after using GridSearchCV

Question

I am using Python 3.6.5 and scikit-learn 0.23.2

from sklearn.linear_model import Ridge
from sklearn.model_selection import GridSearchCV, cross_val_score

ridge = Ridge()

r_parameters = {'alpha':[1e-15, 1e-10, 1e-8, 1e-4, 1e-3, 1e-2, 1, 5, 10, 20]} # this is the Ridge regressor penalty, across different values

ridge_regressor = GridSearchCV(ridge, r_parameters, scoring = 'neg_mean_squared_error', cv = 5)

ridge_regressor.fit(X,y)

ridge_best_params_ = ridge_regressor.best_params_
ridge_best_score_ = -ridge_regressor.best_score_

This has successfully provided me the best_params_ and best_score_ values. Meaning .fit has ran. refit has not been adjusted, and hence it should be default refit = True

However when it comes to trying to return the coefficients of my ridge regression model:

for coef, col in enumerate(X.columns):
    print(f"{col}:  {ridge.coef_[coef]}")

It leads me to the error below:

AttributeError                            Traceback (most recent call last)
<ipython-input-7-e59d1af522dc> in <module>
      2 
      3 for coef, col in enumerate(X.columns):
----> 4     print(f"{col}:  {ridge.coef_[coef]}")

AttributeError: 'Ridge' object has no attribute 'coef_

Appreciate any help for this.

Answer 1

ridge (the Ridge instance) doesn't actually get fitted when fitting ridge_regressor (the GridSearchCV instance); instead, clones of ridge are fitted, and one such is saved as ridge.best_estimator_. So ridge.best_estimator_.coef_ will contain the refitted model's coefficients.

Note that GridSearchCV does provide some convenience functions for accessing the best_estimator_; e.g. ridge.score is just shorthand for ridge.best_estimator_.score, and similarly for, predict and its variants. But it doesn't provide this kind of passthrough for all of the methods/attributes of best_estimator_, and coef_ is one of those not available.