Computer scienceData scienceMachine learningEnsemble learning

Gradient boosting

Setting the learning rate

Report a typo

Select the correct statements about the $\alpha$ values in gradient boosting.

Select one or more options from the list

There is a tradeoff between the number of boosting rounds and the learning rate

Higher learning rates might lead to overfitting, but the convergence will be faster

\alpha \in (0,1]

is a hyperparameter that needs to be tuned manually, with 0.1 being a typical starting value

Higher learning rates are less sensitive to the noise in the data

Setting a lower learning rate might improve the generalization

___

Create a free account to access the full topic