[ADMB Users] Are parameters scaled for optimization.

[John Sibert]

Yes, all init_ parameters are scaled relative to their upper and lower 
bounds, but I'm not sure what the scaled range is, or what happens if 
not bounds are specified.

All of the relevant functions can be found in linad99/boundfun.cpp. 
Anyone interested could have a look, write some dox, and answer the 
questions. This is an open source project.

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 01/03/2014 04:53 PM, Mark Maunder wrote:
> I was meaning does ADMB automatically scale the parameters internally to improve optimization. Based on what someone told me is that they are scaled between 0 and 1 with respect to the lower and upper bounds.
>
> Regards,
>
> Mark
> IATTC
>
>> On Jan 3, 2014, at 5:56 PM, "dave fournier" <davef at otter-rsch.com> wrote:
>>
>> OK, I added scaling for df1b2_init_bounded_vectors (init_bounded_vector in RE models).
>>
>>    as well as unbounded vectors.
>>
>> This appeared to only need a few changes in the file df1b2f15.cpp (attached).
>>
>> However I'm not sure what
>>
>>   > Re: [ADMB Users] Are parameters scaled for optimization.
>>
>> means.  First time around I interpreted it as the ability to scale independent variables so that the optimization
>> runs better and so that the condition number of the problem is smaller.
>>
>> However more literally it seems to mean that the parameters scale themselves in some fashion which is not the case.
>>
>> there is a test file test.tpl to test the changes.
>> <df1b2f15.cpp>
>> <test.tpl>
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users
>