Acceptance/Rejection of new Trees with Randomization


Is it possible to reject new trees based on further conditions?

I want to improve the prediction@k for the top 0.x% with a binary classifier using binary:logistic objective. So my idea was to randomize via colsample but to reject those trees that don’t help with improving prediction@k. Similar to proposals in some papers regarding acceptance/rejection of trees for RFs, but for xgboost.

No, XGBoost currently does not offer such capability.