Function LoadPools

catboost/libs/train_lib/train_model.cpp:95–149

static TDataProviders LoadPools(
    const NCatboostOptions::TPoolLoadParams& loadOptions,
    ETaskType taskType,
    ui64 cpuRamLimit,
    EObjectsOrder objectsOrder,
    TDatasetSubset trainDatasetSubset,
    TConstArrayRef<TDatasetSubset> testDatasetSubsets,
    bool forceUnitAutoPairWeights,
    TVector<NJson::TJsonValue>* classLabels,
    NPar::ILocalExecutor* const executor,
    TProfileInfo* profile
) {
    const auto& cvParams = loadOptions.CvParams;
    const bool cvMode = cvParams.FoldCount != 0;
    CB_ENSURE(
        !cvMode || loadOptions.TestSetPaths.empty(),
        "Test files are not supported in cross-validation mode"
    );

    auto pools = NCB::ReadTrainDatasets(
        taskType,
        loadOptions,
        objectsOrder,
        !cvMode,
        trainDatasetSubset,
        testDatasetSubsets,
        forceUnitAutoPairWeights,
        classLabels,
        executor,
        profile);

    if (cvMode) {
        if (cvParams.Shuffle && (pools.Learn->ObjectsData->GetOrder() != EObjectsOrder::RandomShuffled)) {
            TRestorableFastRng64 rand(cvParams.PartitionRandSeed);

            auto objectsGroupingSubset = NCB::Shuffle(pools.Learn->ObjectsGrouping, 1, &rand);
            pools.Learn = pools.Learn->GetSubset(objectsGroupingSubset, cpuRamLimit, executor);
        }

        TVector<TDataProviders> foldPools = PrepareCvFolds<TDataProviders>(
            std::move(pools.Learn),
            cvParams,
            cvParams.FoldIdx,
            /* oldCvStyleSplit */ true,
            cpuRamLimit,
            executor);
        CB_ENSURE(foldPools.size() == 1, "In cross-validation mode, only one fold is supported");

        profile->AddOperation("Build cv pools");

        return foldPools[0];
    } else {
        return pools;
    }
}

static bool HasInvalidValues(const TVector<TVector<double>>& treeLeafValues) {
    for (const auto& leafValuesDimension : treeLeafValues) {