Skip to content
Quantoros

Why does histogram-based boosting often train faster than searching all distinct continuous split values?

Classic MLLevel: IntermediateHistogram split search

Why does histogram-based boosting often train faster than searching all distinct continuous split values?

Hints
  • Compare the number of bins with distinct observations.
  • Speed comes partly from a discretized representation.

Answer

Features are binned, so split statistics are aggregated over a bounded number of bins rather than every raw value. Binning reduces computation and storage but approximates candidate thresholds.

Source and licence: scikit-learn 1.7.2 — ensemble · BSD-3-Clause · answer: Quantoros editorial

Related questions