factorize.alternate: factorize.alternate

In lutsik/MeDeCom: Decomposition of heterogeneous methylomes

factorize.alternate

Description

Matrix factorization algorithms based on the alternating optimization scheme

Usage

factorize.alternate(
  D,
  k,
  method = "MeDeCom.quadPen",
  t.method = "quadPen",
  Tfix = NULL,
  Tpartial = NULL,
  Tpartial.rows = NULL,
  Apartial = NULL,
  Apartial.cols = NULL,
  V = NULL,
  lambda = 0,
  init = "random",
  opt = 5,
  emp.dim = 500,
  emp.resample = TRUE,
  emp.vsf = 1,
  emp.borders = c(0, 1),
  qp.rangeT = c(0, 1),
  qp.Alower = NULL,
  qp.Aupper = NULL,
  itermax = 100,
  trace = FALSE,
  eps = 1e-08,
  ncores = 1,
  pheno = NULL,
  na.values = FALSE,
  seed = NULL,
  verbosity = 0L
)

Arguments

D	m by n input matrix with mixture data
k	number of latent components, integer
method	optimization method used. Currently supported values are "MeDeCom.quadPen" and "MeDeCom.cppTAfact".
t.method	method for updating the latent component matrix, one of "integer", "empirical", "Hlasso" or "quadPen"
Tfix	an optional matrix of a priori known fixed components
V	for t.method "integer" a small vector of possible values; for t.method "empirical" a vector giving empirical distribution of T values
init	type of initialization, either "random" (default) or "fixed".
opt	if init is "random" number of runs with independent initialization, if init is "fixed" starting values for T and A (see details)
emp.dim	for t.method "empirical", number of randomly drawn samples for T row selection
emp.resample	for t.method "empirical", a flag indicating whether resampling should be done at each iteration
itermax	maximal number of iterations
trace	a flag indicating whether to return the factorization results for each iteration
eps	threshold for objective value change
ncores	number of CPU cores used for parallelization
pheno	a list with phenotypic information
verbosity	flag specifying whether to show diagnostic statements during the execution

Details

In case init is "fixed" the starting values for the m by k matrix of latent components and for the k by n matrix of mixing proportions should be specified as T and A elements of a list supplied as opt

Value

a list with the following elements:

T

matrix of latent components

A

matrix of mixture proportions

Fval

the final value of the objective function

Conv

sequence of objective function values attained after each iteration

rmse

RMSE of the factorization

Author(s)

Martin Slawski

R port by Pavlo Lutsik

lutsik/MeDeCom documentation built on July 15, 2024, 4:51 a.m.