Default Tree Booster Parameters


#1

Hi,

Firstly, thank you for this great work!

I’m an R user and I’d like to double check the default parameters for the Tree Booster. The documentation says default lambda value is 1 for Tree Booster and 0 for Linear Booster. Is this correct?

https://xgboost.readthedocs.io/en/latest/parameter.html#parameters-for-tree-booster

Sorry if this is a silly question. From what I understand, I could see no reason why these two would be different.

Also, there is a get_params function in the python implementation, is this in the R package?

Kind regards,

Oliver


#2

I could see no reason why these two would be different

For linear booster, the resulting model is a linear regression model (y = dot(beta, x), where beta is coefficients). Usually it is not necessary to put regularization penalty on beta, since the linear regression model has limited flexibility so it’s unlikely to overfit. On the other hand, decision trees have greater tendency to overfit due to much greater flexibility.