Skip to contents

[Experimental] GPU equivalent of bixverse::rs_mc_scenic. Assumes the sparse data is pre-filtered for the genes / cells to include. Indices are 0-indexed. GBM is not supported on GPU; use the CPU version.

Usage

rs_mc_scenic_gpu(
  sparse_data,
  tf_indices,
  scenic_params,
  wave_byte_budget,
  seed,
  verbose
)

Arguments

sparse_data

Named list with data, indptr, indices, nrow, ncol, format.

tf_indices

Integer vector. 0-indexed positions of TF predictors.

scenic_params

Named list. See bixverse::params_scenic().

wave_byte_budget

Double. VRAM ceiling for per-wave histogram + cumulative tensors (bytes). Default 4 GiB.

seed

Integer. Random seed.

verbose

Integer. 0L - quiet; 1L - normal; 2L - detailed.

Value

A gene x TF importance matrix.