lnmf: Linked non-negative matrix factorization
In zdebruine/RcppML: Rcpp Machine Learning Library

lnmf	R Documentation

Linked non-negative matrix factorization

Run lNMF on a list of datasets to separate shared and unique signals.

lnmf(
  data,
  k_wh,
  k_uv,
  tol = 1e-04,
  maxit = 100,
  L1 = c(0, 0),
  L2 = c(0, 0),
  seed = NULL,
  mask = NULL
)

`data`	list of dense or sparse matrices giving features in rows and samples in columns. Rows in all matrices must correspond exactly. Prefer `matrix` or `Matrix::dgCMatrix`, respectively
`k_wh`	rank of the shared factorization
`k_uv`	ranks of the unique factorizations, an array corresponding to each dataset provided
`tol`	tolerance of the fit
`maxit`	maximum number of fitting iterations
`L1`	LASSO penalties in the range (0, 1], single value or array of length two for `c(w & u, h & v)`
`L2`	Ridge penalties greater than zero, single value or array of length two for `c(w & u, h & v)`
`seed`	single initialization seed or array of seeds. If multiple seeds are provided, the model with least mean squared error is returned.
`mask`	list of dense or sparse matrices indicating values in `data` to handle as missing. Prefer `Matrix::ngCMatrix`. Alternatively, specify a string "`zeros`" or "`NA`" to mask either all zeros or NA values.

Detailed documentation to come

an object of class lnmf

Zach DeBruine

DeBruine, ZJ, Melcher, K, and Triche, TJ. (2021). "High-performance non-negative matrix factorization for large single-cell data." BioRXiv.