Function GetSubsetForBuildBorders

catboost/libs/data/quantization.cpp:118–142

TODO(akhropov): maybe use different sample selection logic for sparse data

    // TODO(akhropov): maybe use different sample selection logic for sparse data
    static TSubsetIndexingForBuildBorders GetSubsetForBuildBorders(
        const TFeaturesArraySubsetIndexing& srcIndexing,
        const TFeaturesLayout& featuresLayoutForQuantization,
        const TQuantizedFeaturesInfo& quantizedFeaturesInfo,
        EObjectsOrder srcObjectsOrder,
        const TQuantizationOptions& options,
        TRestorableFastRng64* rand,
        NPar::ILocalExecutor* localExecutor
    ) {
        if (NeedToCalcBorders(featuresLayoutForQuantization, quantizedFeaturesInfo)) {
            TFeaturesArraySubsetIndexing subsetIndexing = GetArraySubsetForBuildBorders(
                srcIndexing.Size(),
                /*TODO(kirillovs): iterate through all per feature binarization settings and select smallest
                 * sample size
                 */
                quantizedFeaturesInfo.GetFloatFeatureBinarization(Max<ui32>()).BorderSelectionType,
                srcObjectsOrder == EObjectsOrder::RandomShuffled,
                options.MaxSubsetSizeForBuildBordersAlgorithms,
                rand
            );
            return TSubsetIndexingForBuildBorders(srcIndexing, subsetIndexing, localExecutor);
        } else {
            return TSubsetIndexingForBuildBorders();
        }
    }

    template <class TColumn>
    static ui32 GetNonDefaultValuesCount(const TColumn& srcFeature) {