default value for n.trees in predict()

[AMBarbosa]

predict.GBMFit() requires explicitly specifying n.trees. Would it not be a good idea to provide n.trees with a (reasonable) default value, like length(trees(object)) or the same default as in gbm::predict.gbm()? Cheers

[gregridgeway]

The value of n.trees when predicting should come from doing cross-validation. I don’t know of any other way to pick a default value. As a generally habit, I do not rely on defaults. I’d rather intentionally pick the specific values that make the most sense for the application. The (shrinkage, n.trees) combination depends a lot on the specific dataset. From: AMBarbosa ***@***.***> Sent: Wednesday, August 7, 2024 5:20 PM To: gbm-developers/gbm3 ***@***.***> Cc: Subscribed ***@***.***> Subject: [gbm-developers/gbm3] default value for n.trees in predict() (Issue #166) predict.GBMFit() requires explicitly specifying n.trees. Would it not be a good idea to provide n.trees with a (reasonable) default value, like length(trees(object)) or the same default as in gbm::predict.gbm()? Cheers — Reply to this email directly, view it on GitHub <#166> , or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACERTQDVHKBTL5XZ2FAAV4DZQKFRTAVCNFSM6AAAAABMFGOZR2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGQ2TIMZXGAZTSNY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>