Run geosketch to subsample a matrix — geosketch • sketchR

Perform geometric sketching with the geosketch python package.

geosketch(
  mat,
  N,
  replace = FALSE,
  k = "auto",
  alpha = 0.1,
  seed = NULL,
  max_iter = 200,
  one_indexed = TRUE,
  verbose = FALSE
)

Arguments

mat: m x n matrix. Samples (the dimension along which to subsample) should be in the rows, features in the columns.
N: Numeric scalar, the number of samples to retain.
replace: Logical scalar, whether to sample with replacement.
k: Numeric scalar or "auto", specifying the number of covering. If k = "auto" (the default), it is set to sqrt(nrow(mat)) for replace = TRUE and to N for replace = FALSE.
alpha: Numeric scalar defining the acceptable interval around k. Binary search halts when it obtains between k * (1 - alpha) and k * (1 + alpha) covering boxes.
seed: Numeric scalar or NULL (default). If not NULL, it will be converted to integer and passed to numpy to seed the random number generator.
max_iter: Numeric scalar giving the maximum iterations at which to terminate binary search in rare cases of non-monotonicity of covering boxes.
one_indexed: Logical scalar, whether to return one-indexed indices.
verbose: Locigal scalar, whether to print logging output while running.

Value

A numeric vector with indices to retain.

Details

The first time this function is run, it will create a conda environment containing the geosketch package. This is done via the basilisk R/Bioconductor package - see the documentation for that package for troubleshooting.

References

Hie et al (2019): Geometric sketching compactly summarizes the single-cell transcriptomic landscape. Cell Systems 8, 483–493.

Author

Charlotte Soneson, Michael Stadler

Examples

x <- matrix(rnorm(500), nrow = 100)
geosketch(mat = x, N = 10, seed = 42)
#>  [1] 10 20 34 36 42 45 50 62 70 95