Function GetNumFeatureValuesSample

catboost/python-package/catboost/helpers.cpp:358–395

}

void GetNumFeatureValuesSample(
    const TFullModel& model,
    const NCatboostOptions::TDatasetReadingParams& datasetReadingParams,
    int threadCount,
    const TVector<ui32>& sampleIndicesVector,
    const TVector<TString>& sampleIdsVector,
    TVector<TArrayRef<float>>* numFeaturesColumns
) {
    auto localExecutor = GetCachedLocalExecutor(threadCount).Get();

    auto sampler = GetProcessor<IDataProviderSampler, TDataProviderSampleParams>(
        datasetReadingParams.PoolPath,
        TDataProviderSampleParams {
            datasetReadingParams,
            /*OnlyFeaturesData*/ true,
            /*CpuUsedRamLimit*/ Max<ui64>(),
            localExecutor
        }
    );

    TDataProviderPtr dataProvider;
    if (!sampleIndicesVector.empty()) {
        dataProvider = sampler->SampleByIndices(sampleIndicesVector);
    } else if (!sampleIdsVector.empty()) {
        dataProvider = sampler->SampleBySampleIds(sampleIdsVector);
    } else {
        CB_ENSURE(false, "Neither indices nor sampleIds are provided");
    }
    auto objectsDataProvider = dataProvider->ObjectsData;
    auto rawObjectsDataProvider = dynamic_cast<const TRawObjectsDataProvider*>(objectsDataProvider.Get());
    CB_ENSURE(rawObjectsDataProvider, "Only non-quantized datasets are supported now");

    for (const auto& floatFeature : model.ModelTrees->GetFloatFeatures()) {
        auto values = (*rawObjectsDataProvider->GetFloatFeature(floatFeature.Position.Index))->ExtractValues(localExecutor);
        auto dst = (*numFeaturesColumns)[floatFeature.Position.FlatIndex];
        Copy(values.begin(), values.end(), dst.begin());
    }
}

TMetricsAndTimeLeftHistory GetTrainingMetrics(const TFullModel& model) {
    if (model.ModelInfo.contains("training"sv)) {