Set multiple parameters from GridSearch at once

Question

After using GridSearchCV on my dataset, I want to extract all the best parameters.

from sklearn.tree import DecisionTreeClassifier

params_grid = {'max_depth': np.arange(3, 10),
              'max_leaf_nodes':list(xrange(20,100,20)),
              'max_features':list(xrange(2,10,2))
               }
my_dt = DecisionTreeClassifier()
grid_clf = GridSearchCV(my_dt, params_grid)
grid_clf.fit(xtrain, ytrain)

best_plist = grid_clf.best_params_

Now I want to create a new a DecisionTreeClassifier object and set it's paramters the same as the best_plist. The problem is that I am doing it manually for each parameter like this :

new_clf = DecisionTreeClassifier()
new_clf.max_depth=bestplit['max_depth']  # and so on for all the parameters

However, I will be analyzing multiple ML algoritihms and would want a generic way of setting the parameters of the new object instead of hard-coding for every algorithm. I just wanted to know is there any work around for this in sklearn for this ?

Answer 1

scikit-learn has native way to do what you are trying here. What you need to do is save your new object as the best combination of parameters in your previous fit. That can be achieved with

new_clf = grid.clf.best_estimator_

Which will create a new learner object with the best configuration in your previous search.

PD: this is an old question and maybe you already figured it out, but nevertheless is unanswered.