Skip to contents

Run randomised SVD over a matrix

rs_random_svd.Rd

[Experimental] Runs a randomised singular value decomposition over a matrix. This implementation is faster than the full SVD on large data sets, with slight loss in precision.

Usage

rs_random_svd(x, scale, rank, seed, oversampling, n_power_iter)

Arguments

x

Numeric matrix. Rows = samples, columns = features.

scale

Boolean. Shall the columns be variance normalised. (Mean centring will automatically occur.)

rank

Integer. The rank to use.

seed

Integer. Random seed for reproducibility.

oversampling

Optional integer. Defaults to 10L if NULL.

n_power_iter

Optional integer. Number of power iterations (each with a QR decomposition). Defaults to 2L if NULL.

Value

A list with:

  • scores - u matrix of the SVD multiplied by the singular values.

  • v - v matrix of the SVD.

  • s - Singular values of the SVD.

  • scaled - Boolean. Was the matrix scaled.