Struct forester::split::BestSplitRandomFeature

pub struct BestSplitRandomFeature { /* fields omitted */ }

Find best split, in a number of randomly selected features.

Normally, exactly n_features are tested. However, there are two notable exceptions:

1. if a "perfect split" is encountered (split criterion is exactly zero) the search stops early.
2. if after no split has been found after n_features the search continues until it finds one or runs out of features.

Methods

impl BestSplitRandomFeature

pub fn new(n_features: usize) -> Self

Trait Implementations

impl SplitFinder for BestSplitRandomFeature

fn find_split<Sample, Training: ?Sized>(     &self,     data: &mut Training ) -> Option<Split<Sample::ThetaSplit, Sample::Feature>> where     Sample: SampleDescription,     Training: TrainingData<Sample>,

Attempt to find a split for the given data set.