Compatibility wrapper for column-wise sparse top-k selection

run_sparse_topk_by_column() preserves the column-wise interface used in earlier releases: only explicitly stored entries participate, and missing ranks are padded with NA indices and zero values. New code should call run_sparse_topk() or run_sparse_topk_stored() to select the intended sparse-zero semantics explicitly.

Usage

run_sparse_topk_by_column(x, k, decreasing = TRUE)

Arguments

x

A dgCMatrix (or something coercible to one).

k

Number of stored entries to retain per column.

decreasing

Whether to sort in decreasing order. Defaults to TRUE.

Value

A list with idx and value matrices. Columns with fewer than k stored entries use NA indices and zero values for the remaining ranks.