vsparse

vsparse provides VCSCArray/VCSRArray, standalone Value-Compressed Sparse Column/Row array types implemented in NumPy and accelerated with Numba. It also provides an optional AnnData integration, but the array types themselves have no dependency on AnnData and can be used on their own.

VCSC/VCSR are compressed-sparse layouts inspired by IVSparse’s VCSC. In addition to the usual compressed-sparse pointer/index arrays, nonzero values within each major-axis slice (columns for VCSC, rows for VCSR) are deduplicated: each unique value is stored once, alongside the list of minor-axis positions that share it. This is a strict memory win whenever values repeat heavily within a slice — as is typical for integer count matrices, e.g. single-cell RNA-seq counts.

• Usage
  • Converting an AnnData object
  • Converting from/to scipy sparse
  • Supported operations
  • VCSCAnnData: X backed directly by a VCSCArray
  • Fast IVCSR loading and normalization
• API reference
  • VCSCArray
  • VCSRArray
  • from_anndata()
  • to_layer()
  • load_and_normalize()
  • VCSCAnnData