Run randomised SVD over a matrix

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, rank, seed, oversampling, n_power_iter)

Arguments

x

Numeric matrix. Rows = samples, columns = features.

rank

Integer. The rank to use.

seed

Integer. Random seed for reproducibility.

oversampling

Integer. Defaults to 10L if nothing is provided.

n_power_iter

Integer. How often shall the QR decomposition be applied. Defaults to 2L if nothing is provided.

Value

A list with:

• u - u matrix of the SVD.

• v - v matrix of the SVD.

• s - Eigenvalues of the SVD.