[ADMB Users] mcmc and -noinit

Allan Hicks allan.hicks at noaa.gov
Wed Mar 21 15:50:15 PDT 2012


Hi Mollie,

Ian Taylor just came to my office and mentioned that the -mcgrope option
takes in a number between 0 and 1. And, looking at the manual (why wouldn't
I just do that first) on page 2-2, it mentions that 0.05 or 0.1 is best.
 I'll give that a try with my model later tonight, but it may be worth a
try for you.

Thanks to Ian Taylor for explaining this to me,
Allan

On Wed, Mar 21, 2012 at 3:12 PM, Allan Hicks <allan.hicks at noaa.gov> wrote:

> Hi Mollie,
>
> Last weekend I was playing around with MCMC's in ADMB and I found no
> difference when using mcgrope with different values. So, I'm not exactly
> sure how that works. The reason that I was playing with mcgrope was that
> the ADMB MCMC wasn't traversing the wide space seen with another (unnamed)
> model.  However, playing with -mcrb  worked by allowing the MCMC to jump to
> farther areas.  It may be worth a try.
>
> I hope this helps,
> Allan
>
>
> On Wed, Mar 21, 2012 at 2:07 PM, Mollie Brooks <mbrooks at ufl.edu> wrote:
>
>> Thanks Dave. In that case, I think I'll try getting more samples. I might
>> try mcgrope, but I would have to first make sure that I could explain it in
>> a publication-worthy way.
>> Mollie
>>
>>
>>
>> On 21 Mar 2012, at 3:04 PM, dave fournier wrote:
>>
>> > To the extent that all is working the results should be identical. I
>> checked this with a trivial model.
>> > For a real world example there could be differences as at some point a
>> candidate point on the
>> > boundary of being accepted/rejected could be rejected/accepted.
>> > _______________________________________________
>> > Users mailing list
>> > 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
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/users/attachments/20120321/d87ef159/attachment.html>


More information about the Users mailing list