PySal: Implementation of Geographically Weighted Matrix Decomposition Algorithms in gwlearn

Standard PCA assumes a globally stationary covariance structure, missing spatial heterogeneity in geographic data. This project implements Geographically Weighted PCA and a suite of geographically weighted matrix decomposition algorithms (RobustGWPCA, GWNMF, GWICA, GWSVD, GWKernelPCA) as first-class citizens in the gwlearn sub-package of PySAL, following Harris, Brunsdon and Charlton (2011) and Harris et al. (2014). Built on libpysal.graph.Graph with a scikit-learn compatible API, the implementation brings Python-native GWPCA: including bandwidth selection, eigenvalue stationarity testing, collinearity diagnostics, and robust multivariate spatial outlier detection, to the scientific Python ecosystem.