Wrap Up Quiz

[teorems]

Hello.

Why should we use KFold CV when comparing the performances of the random forest model vs. the gradient boosting one? I believe Kfold was also used in another occasion before, and i did non understand why. Is it not the basic strategy implemented in cross_validate?

The last question concerning the BalancedBaggingClassifier is not clear, you do not state that the base_estimator should be HistGradient. The default estimator is a decision tree, so I got obviously a worse result.

[glemaitre58]

@teorems

Why should we use KFold CV when comparing the performances of the random forest model vs. the gradient boosting one? I believe Kfold was also used in another occasion before, and i did non understand why. Is it not the basic strategy implemented in cross_validate?

It depends: KFold is the default in cross_validate if the model is a regressor otherwise it will be a StratifiedKFold if the model is a classifier.

@teorems

The last question concerning the BalancedBaggingClassifier is not clear, you do not state that the base_estimator should be HistGradient. The default estimator is a decision tree, so I got obviously a worse result.

True, we could be more explicit to mention that the base estimator is the same gbdt.

[teorems]

Ok, thanks! And so in which cases do we have to use an ad hoc cross validation instead of the default one?

[glemaitre58]

It is the topic of module 7 → Choice of cross-validation

[lesteve]

The ensemble wrap-up has been reworked (different dataset)