Q5: 2 models with None can be better

M3. Hyperparameter tuning M3. Wrap-up quiz 3
#1

In the 3 other statements we compare split-wise but not for this statement I guess. Because for 2 folds models without preprocessors performed better:

quantile_model_scores = results.iloc[11][cv_score_columns]
none_model_scores = results.iloc[12][cv_score_columns]
sum(none_model_scores > quantile_model_scores)
# outputs 2

I completely agree that is always better to use preprocessing but here 2 folds belie this.

Maybe precise in the question “on average” or “accross the folds” ?

By the way, thanks for this very interesting mooc :slight_smile:

#2

(Related to this forum post. )

You are right, either using the ranking of average scores or the fold-to-fold criteria leads to the same conclusion. Maybe the word “always” in option b) is somewhat misleading. Thanks for the feedback!