[ADMB Users] matrix inverse

[dave fournier]

Mollie Brooks wrote:

I guess easiest is if you send me your code and whatever I need to run 
the model.
Then I can run it on my version of ADMB and see what happens.

          Dave
> Thanks Dave.
> The original question was about multivariate normally distributed 
> data. Most entries of my variance-covariance matrix are 0. Will your 
> code for finding the inverse of a sparse Hessian work in this case?
>
> my original question:
> "I need to find the maximum likelihood estimates of a nonlinear model 
> with 10 parameters. The data is multivariate normal. I've written some 
> of the code, but then realized that I wrote it terribly inefficiently. 
> I'm wondering, will ADMB have trouble working with a vector of length 
> 7374 and a sparse variance covariance matrix that is 7374x7374? To get 
> the likelihhod, it will have to invert the matrix. Is there a special 
> function to do this?"
>
> thanks,
> Mollie
>
> Mollie Brooks
> Ph.D. Candidate
> NSF IGERT Fellow
> Biology Department
> University of Florida
> mbrooks at ufl.edu <mailto:mbrooks at ufl.edu>
> www.zoology.ufl.edu/mbrooks <http://www.zoology.ufl.edu/mbrooks>
>
>
>
> On Sep 2, 2010, at 10:31 AM, dave fournier wrote:
>
>> Not quite sure what the original question was, but I
>> assume it was about an RE model with a big sparse Hessian.
>>
>> For the spatial model example the covariance matrix is 10,000x10000
>> which is why the current code needs something like 8GB memory.
>> My new code reduces this considerably by not forming the inverse of the
>> sparse Hessian. Hopefully it can be included into the new code base.
>> As I recall it needed something like 1.5GB memory for a sparse
>> 250^2 x 250^2 Hessian.
>> _______________________________________________
>> Users mailing list
>> Users at admb-project.org <mailto:Users at admb-project.org>
>> http://lists.admb-project.org/mailman/listinfo/users
>>
>