Installation

From PyPI (recommended)

pip install sparse-nmf

With the visualization extras (matplotlib, seaborn):

pip install "sparse-nmf[viz]"

From source

git clone https://github.com/bschilder/sparseNMF.git
cd sparseNMF
pip install -e ".[dev]"

GPU support

sparseNMF requires PyTorch. To run on a GPU, install a CUDA-enabled build of torch before installing this package, otherwise pip will pull the CPU-only wheel from PyPI.

pip install --index-url https://download.pytorch.org/whl/cu124 torch
pip install sparse-nmf

CPU-only is fully supported (correctness is unchanged; throughput drops by ~10-50× depending on shape).

Container

A pinned CUDA image is published to GHCR on every release:

docker pull ghcr.io/bschilder/sparsenmf:latest
docker run --gpus all --rm -it ghcr.io/bschilder/sparsenmf:latest python