Function BestSplit

library/cpp/grid_creator/binarization.cpp:74–112

namespace NSplitSelection {

    TQuantization BestSplit(
        TFeatureValues&& features,
        bool featureValuesMayContainNans,
        int maxBordersCount,
        EBorderSelectionType type,
        TMaybe<float> quantizedDefaultBinFraction,
        const TMaybe<TVector<float>>& initialBorders
    ) {
        if (features.DefaultValue && IsNan(features.DefaultValue->Value)) {
            if (featureValuesMayContainNans) {
                features.DefaultValue = Nothing();
            } else {
                throw (yexception() << "Unexpected Nan value.");
            }
        }

        auto firstNanPos = std::remove_if(features.Values.begin(), features.Values.end(), IsNan);
        if (firstNanPos != features.Values.end()) {
            if (featureValuesMayContainNans) {
                features.Values.erase(firstNanPos, features.Values.end());
            } else {
                throw (yexception() << "Unexpected Nan value.");
            }
        }

        if (quantizedDefaultBinFraction) {
            Y_ENSURE(
                (*quantizedDefaultBinFraction >= 0.0f) && (*quantizedDefaultBinFraction < 1.0f),
                LabeledOutput(quantizedDefaultBinFraction) << " is not in required [0, 1) bounds"
            );
        }

        if (features.Values.empty()) {
            return {};
        }

        const auto binarizer = MakeBinarizer(type);
        return binarizer->BestSplit(std::move(features), maxBordersCount, quantizedDefaultBinFraction, initialBorders);
    }

    THolder<IBinarizer> MakeBinarizer(const EBorderSelectionType type) {
        switch (type) {