Package: RcppML Type: Package Title: Fast Non-Negative Matrix Factorization and Divisive Clustering Version: 1.0.0 Date: 2026-11-13 Authors@R: c( person("Zachary", "DeBruine", email = "zacharydebruine@gmail.com", role = c("aut", "cre"), comment = c(ORCID = "0000-0003-2234-4827")), person("Markus", "Buehren", role = c("ctb", "cph"), comment = "Hungarian algorithm implementation in inst/include/RcppHungarian.h"), person("Cong", "Ma", role = c("ctb", "cph"), comment = "C++ port of the Hungarian algorithm in inst/include/RcppHungarian.h"), person("Justin", "Silverman", role = c("ctb", "cph"), comment = "Single-header adaptation RcppHungarian.h")) Description: High-performance non-negative matrix factorization (NMF), singular value decomposition (SVD), and divisive clustering for large sparse and dense matrices. Implements alternating least squares with coordinate descent and Cholesky NNLS solvers, diagonal scaling for interpretable factors, cross-validation for automatic rank selection, multiple distribution-based losses (Gaussian, Poisson, Generalized Poisson, Negative Binomial, Gamma, Inverse Gaussian, Tweedie) via iteratively reweighted least squares, regularization (L1, L2, L21, angular, graph Laplacian), and optional GPU acceleration via CUDA. Includes divisive clustering via recursive rank-2 factorization, consensus clustering, and the StreamPress compressed sparse matrix format. Methods are described in DeBruine, Melcher, and Triche (2021) . Depends: R (>= 4.1.0), Matrix License: GPL (>= 3) Copyright: See the file inst/COPYRIGHTS for the copyright holders of bundled third-party code. Imports: Rcpp, graphics, grDevices, methods, stats, tools, utils LinkingTo: Rcpp, RcppEigen RoxygenNote: 7.3.3 Suggests: ggplot2, hdf5r, jsonlite, knitr, pheatmap, randomForest, RColorBrewer, rmarkdown, patchwork, plotly, scales, testthat (>= 3.0.0), uwot, viridis VignetteBuilder: knitr Config/testthat/edition: 3 LazyData: true LazyDataCompression: xz URL: https://github.com/zdebruine/RcppML BugReports: https://github.com/zdebruine/RcppML/issues SystemRequirements: C++17, CUDA Toolkit >= 11.0 (optional, for GPU acceleration) Encoding: UTF-8 Config/pak/sysreqs: nvidia-cuda-dev Repository: https://zdebruine.r-universe.dev Date/Publication: 2026-11-13 20:39:17 UTC RemoteUrl: https://github.com/zdebruine/rcppml RemoteRef: HEAD RemoteSha: bd51784f97c0be54f7719fe2b34f61d2fe6f74b2 NeedsCompilation: yes Packaged: 2026-11-16 14:07:52 UTC; root Author: Zachary DeBruine [aut, cre] (ORCID: ), Markus Buehren [ctb, cph] (Hungarian algorithm implementation in inst/include/RcppHungarian.h), Cong Ma [ctb, cph] (C++ port of the Hungarian algorithm in inst/include/RcppHungarian.h), Justin Silverman [ctb, cph] (Single-header adaptation RcppHungarian.h) Maintainer: Zachary DeBruine