search

gbm-developers / gbm3

Gradient boosted models
133 stars 117 forks source link

[Question] How are split points chosen for regression trees? #134

Closed ben519 closed 6 months ago

ben519 commented 7 years ago

Suppose I choose my objective as "laplace" to minimize absolute loss. How are tree split points chosen? Does gbm search for the split that reduces mae the most or does gbm always use mse to select split points?

gregridgeway commented 7 years ago

Always MSE, but the form of the working response changes depending on the distribution you choose.

On Dec 18, 2016 6:49 PM, "Ben" notifications@github.com wrote:

Suppose I choose my objective as "laplace" to minimize absolute loss. How are tree split points chosen? Does gbm search for the split that reduces mae the most or does gbm always use mse to select split points?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/gbm-developers/gbm/issues/134, or mute the thread https://github.com/notifications/unsubscribe-auth/AIkZwN5tMD0IjQmHP9MBWrYXUGGbXlEAks5rJcacgaJpZM4LQRan .

ben519 commented 7 years ago

Thanks Greg. That would've been my guess, but I wasn't 100% sure since other packages offer multiple splitting criteria. Related - if you could take a look at my question here I would really appreciate it.