volnmf_logdet {vrnmf}

Update volume-regularized matrix R using logdet volume approximation.

Description

volnmf_logdet finds matrix R that minimizes objective ||X-C*R||^2 + w.vol*log(det(R)+delta).

Usage

volnmf_logdet(
  C,
  X,
  R,
  R.constraint = "pos",
  majorate = FALSE,
  extrapolate = TRUE,
  qmax = 100,
  w.vol = 0.1,
  delta = 1,
  err.cut = 0.001,
  n.iter = 1000
)

Arguments

C

Numeric Matrices. Matrices involved in objective function.Matrix R serves as initialization.

X

Numeric Matrices. Matrices involved in objective function.Matrix R serves as initialization.

R

Numeric Matrices. Matrices involved in objective function.Matrix R serves as initialization.

R.constraint

A character. Set up ('pos') or not ('no') non-negative constraints on matrix R (by default 'pos').

majorate

A boolean. Majorate logdet each iteration (by default FALSE).

extrapolate

A boolean. Use Nesterov acceleration (by default FALSE, currently is not supported).

qmax

A numeric. Maximum asymptotic (1 - 1/qmax) of extrapolation step.

w.vol

A numeric. Volume (logdet) weight in objective function.

delta

A numeric. Determinant pseudocount in objective function.

err.cut

A numeric. Stop algorithm if relative erro in R between iteration is less than err.cut.

n.iter

An integer. Number of iterations.

Value

An updated matrix R.