Available on crate feature
alloc only.Expand description
Histogram-based feature binning for streaming tree construction.
Streaming trees need to evaluate potential splits without storing raw data. Histograms discretize continuous features into bins and accumulate gradient/hessian statistics per bin, enabling efficient split evaluation.
Modules§
- bins
- Feature histogram storage with gradient/hessian sums per bin.
- categorical
- Categorical feature binning – one bin per observed category value.
- quantile
- Greenwald-Khanna streaming quantile sketch for bin edge computation.
- uniform
- Equal-width (uniform) binning strategy.
Structs§
- BinEdges
- Bin edge boundaries computed by a binning strategy.
Enums§
- Binner
Kind - Concrete binning strategy enum, eliminating
Box<dyn BinningStrategy>heap allocations per feature per leaf.
Traits§
- Binning
Strategy - A strategy for computing histogram bin edges from a stream of values.