[ADMB Users] Linear multivariate state-space model.

John Sibert sibert at hawaii.edu
Wed Jul 3 17:44:23 PDT 2013


Hi Dan,
There are some usefule materials to get started with state-space models 
in ADMB. Chapter 8 of the ADMB manual describes implementation of a 
simple Kalman filter. Anders Nielsen and I have developed various 
state-space models for tracking data 
http://www.soest.hawaii.edu/PFRP/reprints/Nielsen_Sibert_2007.pdf.
Regards,
John Sibert

John Sibert
Emeritus Researcher, SOEST
University of Hawaii at Manoa
Honolulu HI (GMT-10)
808-294-3842

Visit the ADMB project http://admb-project.org/

On 07/03/2013 08:15 AM, H. Skaug wrote:
> Hi Dan,
>
> Since your model is linear there are at least two approaches you can take:
> 1. Implement the Kalman filter (see chapter 8 in the ADMB manual)
> 2. Use the random effects module of ADMB in which you explicitly define
>     your state variable X to be a latent variable. This approach
>     can be generalized to nonlinear models.
>
> I recommend doing both as an exercise to understand the possibilities 
> that ADMB offers. With regard to 2) there are a few tricks that people 
> have tpl-files for.
>
> What is the size of your model, i.e. the dimension of X and the length 
> of your
> time series? This info is critical for assessing which of 1 and 2 (if 
> any) is
> the best choice. You say "spatio" which indicates that X is high 
> dimentional.
>
> Hans
>
>
> On Wed, Jul 3, 2013 at 6:34 PM, Dan Gladish <dan.gladish at gmail.com 
> <mailto:dan.gladish at gmail.com>> wrote:
>
>     Hello all,
>
>     I am currently working on a spatio-temporal model that we are
>     hoping to implement in ADMB. Unfortunately, I am very new at ADMB
>     and am looking for a little help.  Specifically, I am looking to
>     see if there are any examples of code for a multivariate linear
>     state-space model:
>
>     Y(t) = X(t) + e(t), e(t) \sim N(0,s2e I)
>     X(t) = M*X(t-1) + eta(t), eta(t) \sim N(0,Q)
>
>     where Y(t) is the data vector, X(t) is the latent variable vector,
>     M is the propagator matrix, and s2e and Q the innovation variance
>     and latent covariance, respectively.
>
>     There are a few examples of state-space models, but I did not see
>     one using a very bare bones set-up like this.  Any help would be
>     much appreciated for a lowly beginner like me.
>
>     Sincerely,
>     Dan Gladish
>
>     _______________________________________________
>     Users mailing list
>     Users at admb-project.org <mailto:Users at admb-project.org>
>     http://lists.admb-project.org/mailman/listinfo/users
>
>
>
>
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users




More information about the Users mailing list