mims-harvard/nimfa

Implements multiple NMF algorithms (Lee-Seung, NMF with projected gradients, probabilistic approaches) with flexible initialization strategies and built-in evaluation metrics (reconstruction error, explained variance, sparseness). Built on NumPy/SciPy for efficient numerical computation, it provides a unified API for comparing factorization methods and integrates with data fusion workflows, particularly for biomedical applications like gene expression analysis and drug discovery.

557 stars. No commits in the last 6 months.