ParOptions

Location: ..\cubatch\Model_PARAFAC\private\ParOptions.m

function Options = ParOptions (varargin);

Description:
Defines the Options for GenPARAFAC and GNPARAFAC

Inputs:
varargin: it can be one structure (in which case the function controls only if the values in the structure are available) or a list of option names and obtions values
Options = ParOptions (optionname1,optionvalue1,optionname2,....);
In the latter case the non specified options will have default value.

Outputs:
Options: Options structure

Field Values Function
cles 'on'/'off' (Default:'on') activate Components Loadings Equal Scaling
compmaxiter # (Default: 3) maximum number of iterations for the compression step
compmode 0 (Default), # which mode is to be compressed (Default "all")
compression 'on'/'off' (Default:'on') activate compression
diagnostics 'on'/'off' (Default:'on') activate the display of the initial diagnostics for the PARAFAC model
display 'none'/# (Default: 1) if 'none' nothing is shown, otherwise the fit and other local diagnostics are shown every # iteration
fitconvcrit # (Default: eps) convergence criterion: relative fit value
frin 'on'/'off' (Default:'on') activate Factor Reparametrisation by Infinite Norm
gradconvcrit # (Default: 10-9) convergence criterion: gradient infinite norm
initialisation 'random'/'orth'/'nipals'
/'dtld'/'swatld'/'best' (Default: 'random')
initialisation method, see InitPar
inititer # (Default: 5) max number of iterations for iterative initialisation methods
inittol # (Default: 10-5) convergence criterion for iterative initialisation method: relative fit decrease
jacsv 'on'/'off' (Default:'off') compute Jacobian Singular Values
lambdaudpar [#1,#2] factor for updating the damping parameter (#1 > 1, 0<#2<1)
lambdaupdate 'levmar'/'hbn'/'alpha' (Default:'levmar') original Levenberg-Marquadt/same as 'levmar' but using a continuous instead o a discreat threshold function/based on the step length(if applicable)
large 'on'/'off' (Default:'off') if 'on' , it uses PARJac, if 'off' it uses PARJacS
linesearch 'none'/'diverg'/'iter' (Default: 'none') no line search/do line-search when divergence occur/do line-search at every iteration
maxiter # (Default: 500) max. number of iterations in fitting
newton 'on'/'off' (Default:'off') if 'on' , the full second derivatives are computed
normeqsolver 'ldivide','pcg','cholesky'
(Default:
'ldivide')
method for solving normal equation system: ldivide/preconditioned conjugate gradients/cholesky and back substitution
parconvcrit # (Default: 10-8) convergence criterion: relative change in the parameters
pcgmaxiter # (Default: 20) max. number of iterations when using 'pcg' as a solver
pcgprecond matrix (Default: []) preconditioner for 'pcg'
pcgtol # (Default: 10-5) tolerance for 'pcg'
refmaxiter # (Default: 10000) max. number of iterations in refining
relfitconvcrit # (Default: 10-6) convergence criterion: relative fit decrease in refining the solution
separate 'on'/'off' (Default:'off') if 'on' , variable separation is employed
tuckextra # (Default: 2) Tucker extra components for compression; the extracted Tucker3 model has dimensions [F+# F+# F+#]


Called by
:
Model_PARAFAC\PARAFACGNFit, Model_PARAFAC\GNParafac, Model_PARAFAC\GenParafac

Subroutines:
Internal: None
External: None

Author:
Giorgio Tomasi
Royal Agricultural and Veterinary University
MLI, LMT, Chemometrics group
Rolighedsvej 30
DK-1958 Frederiksberg C
Danmark

Last modified: 07-Jun-2002 19:16:08

Contact: Giorgio Tomasi, gt@kvl.dk

References