R help - Aug 2006 - gnlsControl

When I run gnls I get the error:

 

Error in nls(y ~ cbind(1, 1/(1 + exp((xmid - x)/exp(lscal)))), data = xy,  :


        step factor 0.000488281 reduced below 'minFactor' of 0.000976563

 

My first thought was to decrease minFactor but gnlsControl does not contain
minFactor nor nlsMinFactor (see below).  It does however contain nlsMaxIter
and nlsTol which I assume are the analogs of maxiter and tol in nls.control.
I would be happy to hear from anyone who has an idea on what parameters in
gnlsControl to change to get convergence.  

 

Cheers

 

Dan Coleman

Genentech Inc.

 
> gnlsControl
function (maxIter = 50, nlsMaxIter = 7, msMaxIter = 50, minScale = 0.001, 

    tolerance = 1e-06, nlsTol = 0.001, msTol = 1e-07, msScale = lmeScale, 

    returnObject = FALSE, msVerbose = FALSE, apVar = TRUE, .relStep
(.Machine$double.eps)^(1/3),

    nlmStepMax = 100, opt = c("nlminb", "optim"),
optimMethod = "BFGS",

    minAbsParApVar = 0.05) 

{

    list(maxIter = maxIter, nlsMaxIter = nlsMaxIter, msMaxIter = msMaxIter, 

        minScale = minScale, tolerance = tolerance, nlsTol = nlsTol, 

        msTol = msTol, msScale = msScale, returnObject = returnObject, 

        msVerbose = msVerbose, apVar = apVar, nlmStepMax = nlmStepMax, 

        opt = match.arg(opt), optimMethod = optimMethod, .relStep .relStep, 

        minAbsParApVar = minAbsParApVar)

}

 
> nls.control
$maxiter

[1] 50

 

$tol

[1] 1e-05

 

$minFactor

[1] 0.0009765625


	[[alternative HTML version deleted]]

Daniel Coleman <coleman.daniel <at> gene.com> writes:
> 
> When I run gnls I get the error:
> 
> Error in nls(y ~ cbind(1, 1/(1 + exp((xmid - x)/exp(lscal)))), data = xy, 
:
> 
>         step factor 0.000488281 reduced below 'minFactor' of
0.000976563
> 
> My first thought was to decrease minFactor but gnlsControl does not contain
> minFactor nor nlsMinFactor (see below).  It does however contain nlsMaxIter
> and nlsTol which I assume are the analogs of maxiter and tol in
nls.control.
> I would be happy to hear from anyone who has an idea on what parameters in
> gnlsControl to change to get convergence.  
Try nlsTol with a large value, e.g. 0.3. If I am really desparate, I put the
gnls in a try() structure, halving nlsTol until it fails, and take the last
successful. 

Dieter