[ADMB Users] a difference in estimates between versions?

Derek Seiple dseiple84 at gmail.com
Wed Oct 5 13:25:26 PDT 2011


Thanks Weihai. That part makes me happy :)
-Derek-


On Wed, Oct 5, 2011 at 3:40 PM, Weihai Liu <liuweih at msu.edu> wrote:
> Hi Derek,
> Sorry for my previous runs which has big gradient and hit the max function
> evaluation and being ignored by me.
> I rerun the ADMB catage example, which agreed on the same results on windows
> 32 bits ADMB 9.0 and latest source built ADMB 10.1, also on ubuntu 64 bits
> latest built. So the problem is coming from his model, there is nothing
> related with ADMB.
> weihai
> On Wed, Oct 5, 2011 at 3:07 PM, Weihai Liu <liuweih at msu.edu> wrote:
>>
>> Hi Derek,
>> 1) Windows32bit with ADMB 9.0 get results
>>  index   name   value      std.dev
>>      1   r   5.0174e-001    7.1722e-002
>>      2   B1  8.9945e+004 1.3305e+004
>>      3   K   9.0813e+004 7.5575e+003
>>      4   lnq -2.1667e+000 1.8993e-001
>> 2) Windows32bit with latest built (this Sept., after subversion 128) get
>> results
>>  index   name   value      std.dev
>>      1   r   4.9970e-001 7.2940e-002
>>      2   B1  8.9504e+004 1.3599e+004
>>      3   K   9.1202e+004 7.9562e+003
>>      4   lnq -2.1695e+000 1.9095e-001
>>
>> 3) Ubuntu 64bit with latest built get hessian not positive definite
>> warning with no std file.
>> 3) Not run for windows 64 bit with latest built yet (on my home computer,
>> if nobody run it, I will add that later).
>> weihai
>>
>> On Wed, Oct 5, 2011 at 2:27 PM, Derek Seiple <dseiple84 at gmail.com> wrote:
>>>
>>> I ran Saang-Yoon's attached example on 32-bit ubuntu with version 9.1,
>>> 10.1, and the latest version with bug fixes. In ALL THREE I got the
>>> following .std file.
>>>
>>>  index   name   value      std dev
>>>     1   r    4.9790e-01 4.4611e-02
>>>     2   B1   8.8262e+04 9.0362e+03
>>>     3   K    9.1796e+04 4.5790e+03
>>>     4   lnq -2.1705e+00 1.4898e-01
>>>
>>>
>>> There was no difference in values amongst the versions. You will note
>>> though that these values are slightly different from the ones he
>>> supplied for ADMB IDE 450-1. This suggests to me that it is either a
>>> 32-bit-64-bit difference or a windows difference.
>>>
>>> Can anyone else run the attached model for 64-bit on linux and windows
>>> or 32-bit on windows for the two versions of ADMB and report the .std
>>> file for comparison?
>>>
>>> Derek
>>>
>>> On Wed, Oct 5, 2011 at 10:28 AM, dave fournier <davef at otter-rsch.com>
>>> wrote:
>>> > It is always interesting to see what kind of posts produce the most
>>> > response.
>>> > certainly none of mine. Anyway
>>> > there are a few standard things you can do to narrow this down.
>>> > First is to take the output from each model and put it into the other
>>> > version and see if the numbers
>>> > stay the same. That verifies that the code produces the same function
>>> > value.
>>> > Also
>>> > you could reduce the convergence criterion as in -crit 1.e-6 to see if
>>> > you
>>> > can get a little closer to the
>>> > actual minimum.
>>> >
>>> > Although it is not relevant to the discussion your approximation to the
>>> > differential equation
>>> > for the population dynamics is gross.  See the semi-implicit version in
>>> > the
>>> > original
>>> > manual for an idea of how to do it right.
>>> > _______________________________________________
>>> > 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
>>>
>>
>
>



More information about the Users mailing list