streampress

StreamPress Python Package

High-performance compressed sparse matrix I/O for Python using the StreamPress .spz format.

Features

• 5–10× compression over raw float32 CSC binary for sparse matrices
• Fast reads: parallelised chunk decompression beats raw CSC even single-threaded
• Column-oriented chunks: random column access without full decompression
• Optional transpose section: fast row access via pre-stored CSC(Aᵀ)
• Row/column metadata: attach obs/var annotation tables
• Dense support: FP16/QUANT8/rANS codecs for dense matrices

Installation

pip install streampress

Quick Start

import scipy.sparse as sp
import streampress as stp

A = sp.random(10000, 500, density=0.05, format="csc", dtype="float64")

stp.st_write(A, "matrix.spz")
B = stp.st_read("matrix.spz")
info = stp.st_info("matrix.spz")

See https://github.com/zdebruine/streampress for full documentation.

This site is open source. Improve this page.