Skip to content

Latest commit

 

History

History
58 lines (46 loc) · 2.51 KB

cran-comments.md

File metadata and controls

58 lines (46 loc) · 2.51 KB

Test environments

  • Local: Red Hat Enterprise Linux 9.7, R 4.5.2, GCC 13.3
  • GitHub Actions: ubuntu-latest (R release, R devel), windows-latest (R release), macos-latest (R release)

R CMD check results

0 errors | 2 warnings | 1 note

  • WARNING 1: checkbashisms script not installed. This is a system tool for checking shell scripts; not a package defect.

  • WARNING 2: qpdf not installed. Used for PDF size reduction checks; all vignettes are HTML-only.

  • NOTE 1 (CRAN incoming feasibility): Package tarball ~9 MB. This is due to:

    • Seven compressed benchmark datasets (6.5 MB installed) used in vignettes and examples covering NMF, SVD, clustering, and recommendation systems. The largest is pbmc3k (3.7 MB, stored as StreamPress-compressed raw bytes).
    • C++ template library headers (2.4 MB installed) for the Eigen-based NNLS solvers, required at compile time by LinkingTo dependents.
    • Eleven pre-built vignettes.
    • SeuratData is listed in Suggests but is not on CRAN; it is used only in optional eval = FALSE vignette examples for Seurat spatial transcriptomics integration.

Reverse dependencies

Checked all CRAN/Bioconductor packages that depend on, import, or link to RcppML 0.3.7:

  • GeneNMF: imports nmf() only — passes with no changes.
  • phytoclass: imports nnls() using the old positional API (nnls(A, b, cd_maxit, cd_tol)). The new nnls() includes a backward-compatibility shim that accepts the old calling convention and emits a deprecation notice. phytoclass will continue to work; we will coordinate with the maintainer to update to the new API.
  • scater (Bioconductor): runtime dependency; does not directly call any RcppML functions via importFrom().
  • miloR (Bioconductor): LinkingTo only; does not import R functions.
  • CARDspa, flashier: Suggests only — no breakage possible.

Major changes since 0.3.7

This is a major version update (0.3.7 → 1.0.0). Key changes:

  • Complete C++ backend rewrite using Eigen template metaprogramming
  • S4 nmf class replacing the previous list output
  • Built-in cross-validation (nmf(..., test_fraction = ...))
  • Multiple distribution-based losses (Gaussian, Generalized Poisson, Negative Binomial, Gamma, Inverse Gaussian, Tweedie)
  • Optional GPU acceleration via CUDA (gracefully disabled when unavailable)
  • SparsePress/StreamPress compressed sparse matrix I/O
  • Factor networks for multi-layer and multi-modal factorization
  • Backward-compatible shim for nnls() old positional API