Calculates sparse PCA for single cell — rs_sc_pca_sparse

Helper function that will calculate sparse PCA without scaling the data. This has the advantage that you avoid creating a large dense matrix due to scaling; however, it has the disadvantage that the first PC will be heavily influenced by average expression. If random_svd is set to FALSE, Lanczos iterations will be used to solve the SVD; if random_svd is set to TRUE, the randomised version will be used with multiplication of the initial sparse matrix with a much smaller random dense matrix, avoiding holding a large dense matrix in memory.

Usage

rs_sc_pca_sparse_gpu(
  f_path_gene,
  f_path_cell,
  no_pcs,
  pca_params,
  cell_indices,
  gene_indices,
  seed,
  verbose
)

Arguments

f_path_gene: String. Path to the counts_genes.bin file.
f_path_cell: String. Path to the counts_cells.bin file. Used if you wish to use the PFlogPF transformation.
no_pcs: Integer. Number of PCs to calculate.
pca_params: Named list. Contains the parameters to use for this PCA run. (Randomised will ignore, as gpu only supports randomised.)
cell_indices: Integer. The cell indices to use. (0-indexed!)
gene_indices: Integer. The gene indices to use. (0-indexed!)
seed: Integer. Random seed for the randomised SVD.
verbose: Integer. 0L - quiet; 1L - normal verbosity; 2L - detailed verbosity.

Value

A list with with the following items

scores - The samples projected on the PCA space (solved via sparse SVD).
loadings - The loadings of the features for the PCA (solved via sparse SVD).
singular_values - The singular values for the PCA (solved via sparse SVD).