R help - Oct 2012 - gam error message: matrix not +ve definite

Hello, 

I'm running a multimodel analysis which involves fitting several GAM models
as implemented in package mgcv.  The issue I'm having is that when I try to
fit my model, gam gives me the following error message: 'Error in
initial.sp(w * X, S, off) : S[[2]] matrix is not +ve definite.' The strange
part of this is that the error message stops my model fitting function when
run on a linux platform, but not on my local windows machine. Ordinarily I
would just run the analysis on my local machine, but I need to run this from
the linux machine to take advantage of the much larger computing capacity.
The version of mgcv(1.7-21) and R (2.15.1) is the same on both machines. 

The data set to which the model if fitted is too large to post here but it
contains 2209 observations, and the model is of the form: gam(Response~Year
+ s(Dayofyear,k=5,bs='cc') + s(Longitude,Latitude,k=4) +
s(Coast_distance,k=4), family=Gamma('log'), data=dat, gamma=1.4)

I realize this is a very particular question, but any help would be really
appreciated.
Thanks,
Dan.



--
View this message in context:
http://r.789695.n4.nabble.com/gam-error-message-matrix-not-ve-definite-tp4645329.html
Sent from the R help mailing list archive at Nabble.com.

On Sun, Oct 7, 2012 at 3:00 PM, garth <dboyce at dal.ca>
wrote:
> Hello,
>
> I'm running a multimodel analysis which involves fitting several GAM
models
> as implemented in package mgcv.  The issue I'm having is that when I
try to
> fit my model, gam gives me the following error message: 'Error in
> initial.sp(w * X, S, off) : S[[2]] matrix is not +ve definite.' The
strange
> part of this is that the error message stops my model fitting function when
> run on a linux platform, but not on my local windows machine. Ordinarily I
> would just run the analysis on my local machine, but I need to run this
from
> the linux machine to take advantage of the much larger computing capacity.
> The version of mgcv(1.7-21) and R (2.15.1) is the same on both machines.
>
> The data set to which the model if fitted is too large to post here but it
> contains 2209 observations, and the model is of the form: gam(Response~Year
> + s(Dayofyear,k=5,bs='cc') + s(Longitude,Latitude,k=4) +
> s(Coast_distance,k=4), family=Gamma('log'), data=dat, gamma=1.4)
>
> I realize this is a very particular question, but any help would be really
> appreciated.
> Thanks,
> Dan.
>
It's a bit of a shot in the dark, but might it reflect data that is
"close" to not being non-positive-definite and differences in BLAS?

The only other think I'd try would be a

update.packages(checkBuilt = TRUE, ask = FALSE)

to just update every package you've got installed. It might not be
that the error is from mgcv proper, but rather a dependency thereof
which is in different versions between the machines.

Also, I note you're posting through Nabble. Nabble actually only
mirrors the R-Help mailing list, so we (the majority of respondents
who don't use Nabble) don't get a forum view, but rather only emails
of individual posts. The net result of this is that we don't see
context of your follow-up postings unless you specifically include it,
but it's much appreciated if you do. I believe there might also be a
button which does so automatically, but I'll leave finding that to
you.


Cheers,
Michael

Hi Dan,

Any chance that you could send me the data offlist and I'll take a look 
(under the understanding that I'll only use the data for de-bugging of 
course)?

Could you also let me know which linux distribution you are using, 
whether it's 64bit or 32 bit, and, if possible, what BLAS R is using.

best,
Simon

On 07/10/12 15:00, garth wrote:
> Hello,
>
> I'm running a multimodel analysis which involves fitting several GAM
models
> as implemented in package mgcv.  The issue I'm having is that when I
try to
> fit my model, gam gives me the following error message: 'Error in
> initial.sp(w * X, S, off) : S[[2]] matrix is not +ve definite.' The
strange
> part of this is that the error message stops my model fitting function when
> run on a linux platform, but not on my local windows machine. Ordinarily I
> would just run the analysis on my local machine, but I need to run this
from
> the linux machine to take advantage of the much larger computing capacity.
> The version of mgcv(1.7-21) and R (2.15.1) is the same on both machines.
>
> The data set to which the model if fitted is too large to post here but it
> contains 2209 observations, and the model is of the form: gam(Response~Year
> + s(Dayofyear,k=5,bs='cc') + s(Longitude,Latitude,k=4) +
> s(Coast_distance,k=4), family=Gamma('log'), data=dat, gamma=1.4)
>
> I realize this is a very particular question, but any help would be really
> appreciated.
> Thanks,
> Dan.
>
>
>
> --
> View this message in context:
http://r.789695.n4.nabble.com/gam-error-message-matrix-not-ve-definite-tp4645329.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> 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.
>

-- 
Simon Wood, Mathematical Science, University of Bath BA2 7AY UK
+44 (0)1225 386603               http://people.bath.ac.uk/sw283