Open trivialfis opened 2 years ago
We should run https://github.com/dmlc/xgboost/blob/18cbebaeb9c04dfb112b7f42c7b15087bc6a190b/src/common/quantile.cu#L351 before pruning so that accumulated weight for the same element is more accurate.
This might increase the GPU memory usage by a few times. Need better ideas.
We should run https://github.com/dmlc/xgboost/blob/18cbebaeb9c04dfb112b7f42c7b15087bc6a190b/src/common/quantile.cu#L351 before pruning so that accumulated weight for the same element is more accurate.