RcppML: Fast Non-Negative Matrix Factorization and Divisive Clustering

High-performance non-negative matrix factorization (NMF), singular value decomposition (SVD/PCA), and divisive clustering for large sparse and dense matrices, powered by Rcpp and Eigen.

NMF (Non-negative Matrix Factorization)

nmf: Fit NMF model (sparse or dense input, optional cross-validation)
evaluate: Evaluate reconstruction loss of an NMF model
align: Align factors across NMF models
predict,nmf-method: Project new data onto a fitted NMF model
consensus_nmf: Consensus clustering from multiple NMF runs
simulateNMF: Simulate data from a known NMF model
auto_nmf_distribution: Select distribution based on data characteristics

SVD / PCA

svd: Truncated SVD via deflation
pca: PCA (centered SVD)
reconstruct: Reconstruct matrix from SVD/PCA model
variance_explained: Proportion of variance per factor

NNLS (Non-negative Least Squares)

nnls: Solve non-negative least squares problems

Clustering

dclust: Divisive clustering via recursive rank-2 NMF
bipartition: Split samples into two groups via rank-2 NMF
bipartiteMatch: Match two sets of cluster labels

factor_net: Compile a factorization network
fit: Fit a compiled factor network
factor_input, nmf_layer, svd_layer: Node constructors
factor_shared, factor_concat, factor_add: Merge operations
factor_config, W, H: Configuration
cross_validate_graph: Cross-validate a factor network

StreamPress I/O

st_write, st_read: Read/write .spz files
st_info: Inspect .spz file metadata
st_read_obs, st_read_var: Read embedded metadata tables
st_read_gpu, st_free_gpu: GPU-direct .spz reading

GPU

gpu_available: Check GPU availability
gpu_info: Get GPU device details

Utilities

cosine: Cosine similarity
sparsity: Matrix sparsity fraction

Author

Maintainer: Zachary DeBruine zacharydebruine@gmail.com (ORCID)