[ADMB Users] multinomial weights and -gh

H. Skaug hskaug at gmail.com
Fri Mar 1 16:02:33 PST 2013


Hi,

-gh does not work with the multinomial weights in general. That is why
I sent you
a modified df1b2gh.cpp. However, that was for a single random effect.
For multiple random effects -gh is run by a separate code found
in another cpp file. If you develop your code without -gh I can
send you the appropriate file in a while.

Hans

On Sat, Mar 2, 2013 at 12:29 AM, Jeff Laake <jefflaake at gmail.com> wrote:
> So I'm confused. In the binomial example (binomial_w2.tpl) it shows
> multiplication of weight and likelihood value but in the distance sampling
> example I did with Hans the weighting was all behind the scenes to use -gh.
> Can someone hep me with the example cjsre.tpl at:
>
> https://docs.google.com/folder/d/0B77g1ScdUwVeWkNDaFA5OVp2SjA/edit?usp=sharing
>
> which I modeled after what Hans helped me with.  You'll see code multiplying
> weights that is commented out.
>
> The weights here are frequencies of the occurrence of a particular capture
> history.
>
> If you rename the file yes_cjsre.dat to cjsre.dat and run it then it will
> work giving a likelihood about 333.  But if you rename the example
> no_cjsre.dat to cjsre.dat and run it, it will give a value of about 110.
> Both should be in the 300s.  The difference is that in the no_cjsre.dat
> there are two random effects per call  whereas in yes_cjsre.dat it is on
> only one of the parameters. Clearly the weights aren't being applied
> properly when there are random effects on both parameters.
>
> I found the following comment in the code df1bg2h
>
>    // !!! This only works for one random efect in each separable call
>    // at present.
>
> Am I out of luck at present?
>
> --jeff
>
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users
>



More information about the Users mailing list