Setting n_estimators or n_iter_no_change

M6. Ensemble of models M6. Exercise M6.03
#1

Hi,

When reading this notebook, I understand I should always set n_estimators to a big value and try to tune n_iter_no_change.
Is there any advantage to specifically tune n_estimators ?

thanks

#2

I would say that looking for early stopping would be better than tuning n_estimators indeed.

Potentially, you might be interested in a model that should be fast at predicting and therefore not interested in a model that fully converged. In this case, you probably want to monitor the performance of the model for a different number of trees and pick up the model with the minimum number of trees but with a validation loss that is low enough.

1 Like