I am not sure what can i get home from this excersise.
When i read this
We see that the predictor provided by the bagging regressor does not need much hyperparameter tuning compared to a single decision tree.
Should I understand that one does not need to tunnig a bagging regressor?
Could we always be sure that the performance of the baggin regressor will be as good as one tuned via RandomizedSearchCV or some similar function?
Hello @MarioVB,
It is always advised to tune at least the most important hyperparameters of a predictor (by using a RandomizedSearchCV or similar, as you mention).
The paragraph means that the user effort needed for tuning the bagging regressor and the single decision tree are quite similar, as they share their principal hyperparameters. This is not true when tuning n_estimators because it is more costly in terms of computational sources (see discussion here).
Maybe we need some rephrasing here. Thanks for your feedback!
That makes sense to me.
Kind as always, Amor 