get_HDF5_rowVars

C++ Function Reference

1 Signature

Eigen::VectorXd BigDataStatMeth::get_HDF5_rowVars(BigDataStatMeth::hdf5Dataset *dsA, bool bparal, Rcpp::Nullable< int > wsize, Rcpp::Nullable< int > threads)

2 Description

Row variances of an HDF5 matrix (block-wise, parallel).

3 Parameters

  • dsA (BigDataStatMeth::hdf5Dataset *): Open HDF5 dataset.
  • bparal (bool): Enable OpenMP parallelism.
  • wsize (Rcpp::Nullable< int >): Block size (NULL = auto).
  • threads (Rcpp::Nullable< int >): Thread count (NULL = auto).

4 Returns

Vector of length nrows_R.

5 Details

Equivalent to apply(X, 1, var) — uses Bessel’s correction (n-1). If ncol_R == 1 the result is a vector of NaNs, matching base R behaviour.

6 Call Graph

Function dependencies

7 Source Code

File: inst/include/hdf5Algebra/matrixAggregations.hpp • Lines 788-852

inline Eigen::VectorXd get_HDF5_rowVars(BigDataStatMeth::hdf5Dataset* dsA,
                                         bool bparal,
                                         Rcpp::Nullable<int> wsize,
                                         Rcpp::Nullable<int> threads)
{
    try {
        const hsize_t nHDF5rows = dsA->nrows(); // = R ncols  (fixed)
        const hsize_t nHDF5cols = dsA->ncols(); // = R nrows  (iterated)
        const double n = static_cast<double>(nHDF5rows);

        // var undefined for n < 2
        if (nHDF5rows < 2) {
            return Eigen::VectorXd::Constant(nHDF5cols,
                                             std::numeric_limits<double>::quiet_NaN());
        }

        const hsize_t bs = agg_block_size(wsize, nHDF5cols, nHDF5rows);

        std::vector<hsize_t> starts, sizes;
        agg_make_blocks(nHDF5cols, bs, starts, sizes);

        const std::vector<hsize_t> stride = {1, 1}, blk = {1, 1};
        const int nthreads = static_cast<int>(
            BigDataStatMeth::get_threads(bparal, threads));

        Eigen::VectorXd result(nHDF5cols);

        #pragma omp parallel for schedule(dynamic) num_threads(nthreads) \
                shared(dsA, starts, sizes, result)
        for (hsize_t bi = 0; bi < starts.size(); bi++) {
            std::vector<double> vd(nHDF5rows * sizes[bi]);
            //.. 20260325 - remove critical ..// #pragma omp critical(accessFile)
            //.. 20260325 - remove critical ..// { 
                dsA->readDatasetBlock({0, starts[bi]}, {nHDF5rows, sizes[bi]}, stride, blk, vd.data()); 
                //.. 20260325 - remove critical ..// }

            // Map as (ncols_R × block_rrows) RowMajor
            Eigen::Map<const RMMatd> X(vd.data(),
                static_cast<Eigen::Index>(nHDF5rows),
                static_cast<Eigen::Index>(sizes[bi]));

            // Computational formula over R-rows in this block:
            //   var_row = (sum_sq_row - sum_row^2 / n) / (n - 1)
            // where n = ncols_R (HDF5 nrows, all loaded)
            const Eigen::RowVectorXd rowsum   = X.colwise().sum();
            const Eigen::RowVectorXd rowsumsq = X.colwise().squaredNorm();

            result.segment(starts[bi], sizes[bi]) =
                ((rowsumsq.array() - rowsum.array().square() / n) /
                 (n - 1.0)).transpose();
        }

        return result;

    } catch (H5::FileIException& e) {
        throw std::runtime_error("c++ exception get_HDF5_rowVars (File IException): "
                                 + std::string(e.getDetailMsg()));
    } catch (H5::DataSetIException& e) {
        throw std::runtime_error("c++ exception get_HDF5_rowVars (DataSet IException): "
                                 + std::string(e.getDetailMsg()));
    } catch (std::exception& e) {
        throw std::runtime_error(std::string("c++ exception get_HDF5_rowVars: ")
                                 + e.what());
    }
}

8 Usage Example

#include "BigDataStatMeth.hpp"

// Example usage
auto result = get_HDF5_rowVars(...);