What additional randomness distinguishes ExtraTrees splitting from a standard random forest?
Classic MLLevel: IntermediateExtraTrees randomization
What additional randomness distinguishes ExtraTrees splitting from a standard random forest?
Hints
- Random feature subsets are not the only possible randomness.
- Consider how candidate thresholds are proposed.
Answer
Candidate split thresholds are randomly drawn rather than exhaustively optimized over values for each chosen feature. This increases randomization, often reducing variance at the cost of some bias.
Source and licence: scikit-learn 1.7.2 — ensemble · BSD-3-Clause · answer: Quantoros editorial