factor_intensities {vrnmf}

Infer a matrix of non-negative intensities in NMF with offset/nmf-offset.

Description

factor_intensities estimates a non-negative matrix D that optimizes the objective function F = ||X - C*D - offset||^2, where offset is either column-specific offset or a "1-rank nmf term": product of row vector and column vector

Usage

factor_intensities(
  C,
  X,
  fit.nmf = TRUE,
  fit.factor = FALSE,
  qp.exact = FALSE,
  n.iter = 200,
  qp.iter = 10,
  rel.error.cutoff = 1e-05,
  extrapolate = TRUE,
  extrapolate.const = TRUE,
  extrapolate.convex = FALSE,
  q.factor = 1,
  verbose = TRUE,
  n.cores = 1
)

Arguments

C

Numeric matrices.

X

Numeric matrices.

fit.nmf

A boolean. Fit both intensities and spectrum of the offset residuals.

fit.factor

A boolean. Fit only spectrum of the offset residuals (keep intensities constant across samples).

qp.exact

A boolean. Estimate intensities using exact quadratic programming (qp.exact = TRUE) or inexact QP via gradient decent with extrapolation (qp.exact = FALSE).

n.iter

An integer. Number of iterations.

qp.iter

= 1e+1 An integer. Number of iterations of inexact QP.

rel.error.cutoff

A numeric. Relative error cutoff between iterations to stop iterations.

extrapolate

A boolean. Use Nesterov-like extrapolation at each iteration.

extrapolate.const

A boolean. Use extrapolation scheme that adds a constant extrapolation q.factor (described below) at each iteration.

extrapolate.convex

A boolean. Use Nesterov extrapolation scheme.

q.factor

A numeric. Specification of a a constant extrapolation factor used in case of extrapolate.const = T.

verbose

A boolean. Print per-iteration information (by default TRUE).

n.cores

An integer. Number of cores to use.

Value

Fitted matrix D.