Scala spark xgboost (v0.81) incremental training


#1

To Community users and Developers:
I am currently using scala spark xgboost v0.81 and trying to evaluate an existing model’s variable importance on a new data set (not the training data set).

I pretty much like to follow this approach https://github.com/dmlc/xgboost/pull/1670.

However I found scala API has no choice for parameter updater/process_type/refresh_leaf, did I missed something? or some tricks can be played here?

Thanks
Yao


#2

The refresh_leaf option is experimental and thus not available from XGBoost4J-Spark. Also see discussion at https://github.com/dmlc/xgboost/issues/4188