R help - Aug 2010 - maxNR in maxLik package never stops

Greetings, 

 

 I use maxNR function under maxLik package to find the REML estimates of the
parameters of variance components in heteroskedastic linear regression models. I
assume that in the model there is additive/multiplicative/mixed
heteroskedasticity and I need estimate the respective parameters of
additive/multiplicative/mixed variance components.

 

 For my research purposes I make a batch of simulations and estimate for the
given design matrix but different (simulated) response vectors the variance
component parameters. However, it happens very frequently, usually after the
first simulation that the log-likelihood function maximization step (for which
maxNR is used) R "freezes". It appears nothing wrong with R, it is
working but the maximization never stops (frankly, I cannot say that it is
really "never", I did not allow it more time than one day). Therefore,
my simulations crash - I cannot even perform one simulation.

 

 I see two possible solutions:

 1. Using some additional parameters of maxNR (such as number of iterations or
something).

 2. Using some function that could help me control the length of the
maximization step - if it were to take to long, it would be evaluated as an
unsuccessful simulation.

 

 Since I do not have much experience with R, it is clear to me that I cannot
resolve the thing myself. Maybe there are other viable solutions.

 

 I shall be much obliged for your help and advice.

 

 
Martin Boďa 

 (Matej Bel University / Faculty of Natural Sciences / Banská Bystrica /
Slovakia).


   

 
	[[alternative HTML version deleted]]

Hi Martin!

Which version of maxLik do you use?
(You can figure this out, e.g. by typing 'help( package = "maxLik"
)'
at the R prompt.)

If you are not using version 0.8-0, please upgrade to the latest version.

You can get more information on the optimization process by adding
"print.level=4" as argument of maxLik() or maxNR().

You could send us also an example which allows us to replicate your problem.

Best wishes,
Arne


On 29 August 2010 01:06, "Martin Bo?a" <hcd at azet.sk>
wrote:
>
>
>
>
>
>
> Greetings,
>
>
>
> ?I use maxNR function under maxLik package to find the REML estimates of
the parameters of variance components in heteroskedastic linear regression
models. I assume that in the model there is additive/multiplicative/mixed
heteroskedasticity and I need estimate the respective parameters of
additive/multiplicative/mixed variance components.
>
>
>
> ?For my research purposes I make a batch of simulations and estimate for
the given design matrix but different (simulated) response vectors the variance
component parameters. However, it happens very frequently, usually after the
first simulation that the log-likelihood function maximization step (for which
maxNR is used) R "freezes". It appears nothing wrong with R, it is
working but the maximization never stops (frankly, I cannot say that it is
really "never", I did not allow it more time than one day). Therefore,
my simulations crash - I cannot even perform one simulation.
>
>
>
> ?I see two possible solutions:
>
> ?1. Using some additional parameters of maxNR (such as number of iterations
or something).
>
> ?2. Using some function that could help me control the length of the
maximization step - if it were to take to long, it would be evaluated as an
unsuccessful simulation.
>
>
>
> ?Since I do not have much experience with R, it is clear to me that I
cannot resolve the thing myself. Maybe there are other viable solutions.
>
>
>
> ?I shall be much obliged for your help and advice.
>
>
>
>
> Martin Bo?a
>
> ?(Matej Bel University / Faculty of Natural Sciences / Bansk? Bystrica /
Slovakia).
>
>
>
>
>
> ? ? ? ?[[alternative HTML version deleted]]
>
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
>


-- 
Arne Henningsen
http://www.arne-henningsen.name