Handle on a sparse matrix on disk — sparse

Makes a handle on a csr_matrix or csc_matrix Element that is open in an rhdf5 group, as Python's anndata.io.sparse_dataset() does. The handle reads only the parts of data and indices that an index needs (see [, as.matrix() and $to_memory()), and it is also what a backed object holds as X. It reads from the group while the group stays open, and the caller closes the group. handle$append(value) adds the rows (CSR) or columns (CSC) of a sparse matrix in the same format to the file, which must be open read-write and hold resizable data, indices and indptr (as files written by Python do), and the handle's shape grows with it.

Usage

sparse_dataset(group, ..., should_cache_indptr = TRUE)

Arguments

group: An open rhdf5 group handle holding a csr_matrix or csc_matrix Element.
...: Must be empty, because Python takes should_cache_indptr by name only.
should_cache_indptr: Whether to read indptr once and keep it.

Value

A backed_sparse, which has the fields shape, dtype, format ("csr" or "csc") and backend ("hdf5").

Examples

path <- tempfile(fileext = ".h5ad")
x <- Matrix::sparseMatrix(i = c(1, 3), j = c(2, 3), x = c(5, 6))
write_h5ad(AnnData(methods::as(x, "RsparseMatrix")), path)
file <- rhdf5::H5Fopen(path, flags = "H5F_ACC_RDONLY")
group <- rhdf5::H5Gopen(file, "X")
handle <- sparse_dataset(group)
handle$format
#> [1] "csr"
handle[3, ]
#> [1] 0 0 6
rhdf5::H5Gclose(group)
rhdf5::H5Fclose(file)