Boosted tree for regression


#1

Hello i am trying to build a model for regression, please correct me if wrong at the parameters section i should set “objective”:“reg:linear”. My question is: is it making a linear regression model? if it is how can i build a boosted tree for regression (non linear)?? Thanks


#2

“booster”(model) specifies how do you want to represent/approximate the underlying data distribution.
“objective” is the cost function that estimate/measure how far is your model from the data distribution.

I don’t know how to explain it succinctly…


#3

Hello @ars you can set your loss function to be one of the regression losses for the objective parameter.

This means that the trees, which are non-linear learners will be grown using the squared loss error that is also used in linear regression. The parameter naming is a bit unfortunate here, we should probably look into changing this, I’ll create an issue about that.