[ADMB Users] testing hybrid mcmc

[Ian Taylor]

Hi Mollie,
mcsave doesn't work in the hybrid method. This is flagged as Issue #94:
http://www.admb-project.org/redmine/issues/94.
so you just get huge files and have to thin later. If you want to try to
get it working, you can probably start with the commented-out line 384 in
the file hybmcmc.cpp<http://admb-project.org/redmine/projects/issues/repository/entry/trunk/src/nh99/hybmcmc.cpp>
:
    //if ( (on=option_match(ad_comm::argc,ad_comm::argv,"-mcsave"))>-1)

Issue 94 also suggests that it may not work with random effects, so you can
be a pioneer there.

As for the  "-hynstep" (singular, not plural), that input controls how many
steps the algorithm makes between doing something else. If there's a good
explanation out there of how hybrid MCMC works in general, or advise on
tuning the epsilon and n-steps inputs, I haven't seen it.
-Ian


On Fri, Aug 24, 2012 at 11:01 AM, Mollie Brooks <mbrooks at ufl.edu> wrote:

> Hi ADMB users,
>
> This is essentially the simple tpl file that Dave told me to try, but I
> added the random effects vector u.
>
> DATA_SECTION
>    int n
>   !! n=10;
> PARAMETER_SECTION
>    init_vector x(1,n);
>    sdreport_number x1;
>    objective_function_value f;
>    random_effects_vector u(1,n);
> PROCEDURE_SECTION
>    x1=x(1);
>    f+=0.5*norm2(x);
>    f+=0.5*norm2(u);
>
>
> I can compile it, then run it with the command
> ./hybridsimple_re -hybrid -hyeps .3 -hynsteps 2 -mcmc2 10
>
> After playing around for a while, I can't figure out what the parameter
> -hynsteps is doing.
> I also can't seem to use mcsave to thin the samples.
> Anybody know how to use -hynsteps and -mcsave in hybrid mcmc sampling?
>
> thanks,
> Mollie
>
>
> Mollie Brooks
> Postdoctoral Researcher, Ponciano Lab
> Biology Department, University of Florida
> http://people.biology.ufl.edu/mbrooks
>
>
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/users/attachments/20120824/bca25d4d/attachment.html>