Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Support categorical data for hist. (#7695)

Browse Source 
* Extract partitioner from hist.
* Implement categorical data support by passing the gradient index directly into the partitioner.
* Organize/update document.
* Remove code for negative hessian.
This commit is contained in:
Jiaming Yuan
2022-02-25 03:47:14 +08:00
committed by GitHub
parent f60d95b0ba
commit 83a66b4994
15 changed files with 402 additions and 498 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/tree/hist/evaluate_splits.h
View File

@@ -288,10 +288,10 @@ template <typename GradientSumT, typename ExpandEntry> class HistEvaluator {
auto base_weight =
evaluator.CalcWeight(candidate.nid, param_, GradStats{parent_sum});
auto left_weight = evaluator.CalcWeight(
candidate.nid, param_, GradStats{candidate.split.left_sum});
auto right_weight = evaluator.CalcWeight(
candidate.nid, param_, GradStats{candidate.split.right_sum});
auto left_weight =
evaluator.CalcWeight(candidate.nid, param_, GradStats{candidate.split.left_sum});
auto right_weight =
evaluator.CalcWeight(candidate.nid, param_, GradStats{candidate.split.right_sum});
if (candidate.split.is_cat) {
std::vector<uint32_t> split_cats;
@@ -308,11 +308,11 @@ template <typename GradientSumT, typename ExpandEntry> class HistEvaluator {
split_cats = candidate.split.cat_bits;
common::CatBitField cat_bits{split_cats};
}
tree.ExpandCategorical(
candidate.nid, candidate.split.SplitIndex(), split_cats, candidate.split.DefaultLeft(),
base_weight, left_weight, right_weight, candidate.split.loss_chg, parent_sum.GetHess(),
candidate.split.left_sum.GetHess(), candidate.split.right_sum.GetHess());
base_weight, left_weight * param_.learning_rate, right_weight * param_.learning_rate,
candidate.split.loss_chg, parent_sum.GetHess(), candidate.split.left_sum.GetHess(),
candidate.split.right_sum.GetHess());
} else {
tree.ExpandNode(candidate.nid, candidate.split.SplitIndex(), candidate.split.split_value,
candidate.split.DefaultLeft(), base_weight,