Run NMF (HALS) on the GPU over meta cells — rs_nmf_single_mc

GPU counterpart of bixverse::rs_nmf_single_mc(). Assumes that the sparse data is pre-filtered for the cells/genes you wish to include. Indices in the sparse data need to be 0-indexed.

Usage

rs_nmf_single_mc_gpu(
  sparse_data,
  k,
  preprocessing,
  use_second_layer,
  nmf_hals_params,
  seed,
  verbose
)

Arguments

sparse_data: A named list with data, indptr, indices, nrow, ncol and cs_type.
k: Integer. Number of latent factors to return. At most 128, the GPU solver's rank cap.
preprocessing: String. One of c("none", "sd", "sqrt_sd").
use_second_layer: Boolean. If TRUE, runs NMF on normalised counts.
nmf_hals_params: Named list. Contains the NMF parameters.
seed: Integer. Random seed for initialisation.
verbose: Integer. 0L - quiet; 1L - normal verbosity; 2L - detailed verbosity.

Value

A list with w, h, final_loss, n_iter, converged.