Run sparse diffusion map

Performs non-linear dimensionality reduction using a memory-efficient sparse k-NN method. It is better suited for larger datasets than the exact dense method, as its memory usage scales linearly rather than quadratically. Uses an adaptive Gaussian kernel (local sigma based on the distance to the k-th nearest neighbor).

Usage

run_diffusion_sparse(
  data,
  dims = 2,
  k_neighbors = 20,
  oversampling = 10,
  n_iter = 300
)

Arguments

data

Numeric matrix or data.frame. Missing values (NA) and infinite values are not supported and will cause an error.

dims

Integer. Number of diffusion dimensions to return (default 2).

k_neighbors

Integer. Number of nearest neighbors for sparse method (default 20).

oversampling

Integer. Dimensions added for stability in randomized SVD algorithm (default 10). Higher values may negatively impact results of thr algorithm.

n_iter

Integer. Number of subspace iterations in randomized SVD algorithm (default 300).

Value

A diffmap object

Examples

data <- matrix(runif(5000), nrow = 1000, ncol = 5)
model <- run_diffusion_sparse(data, k_neighbors = 20, oversampling = 10, n_iter = 200)