search

Blunde1 / agtboost

Adaptive and automatic gradient boosting computations.
MIT License
66 stars 11 forks source link

passing indices instead of copying data-subsets #22

Closed Blunde1 closed 4 years ago

Blunde1 commented 4 years ago

Old: design matrix X and differentials g and h are copied and subsets are created in memory. New: Keep track of indices of node, and split this instead