Re: large Residual in lsqnonlin
- From: "Beiyan Ou" <beiyanou@xxxxxxx>
- Date: Sun, 11 Dec 2005 17:33:31 -0500
Thanks for your suggestions, John. I'll certainly try using weighted
least squares. Estimation of 5 or more parameters using only one ODE
probably does take a lot of iterations and cpu time.
I face another technical difficulty is that Line-search method does
not handle bound constraints, so if I include lower bound for my
parameters, it switches to large-scale method instead.
John D'Errico wrote:
>
>
> In article <ef1e7c1.0@xxxxxxxxxxxxxxxx>, Beiyan
<beiyanou@xxxxxxx>
> wrote:
>
>> After I changed maxIter to 200, and MaxFunEvals to 2000, the
> program
>> stopped saying:
>> No improvement in search direction: Terminating
>>
>> Any suggestions?
>
> Its very difficult to guess what can be done to
> help here. One possibility is to consider a
> weighted least squares. Simply choose weights
> by which to scale your residuals. This can be
> appropriate if the uncertainty in your model
> increases with residual size. Such proportional
> error is often appropriate.
>
> HTH,
> John D'Errico
>
>
> --
> The best material model of a cat is another, or preferably the
> same, cat.
> A. Rosenblueth, Philosophy of Science, 1945
>
> Those who can't laugh at themselves leave the job to others.
> Anonymous
>
.
- Follow-Ups:
- Re: large Residual in lsqnonlin
- From: John D'Errico
- Re: large Residual in lsqnonlin
- References:
- large Residual in lsqnonlin
- From: Beiyan Ou
- Re: large Residual in lsqnonlin
- From: Beiyan
- Re: large Residual in lsqnonlin
- From: John D'Errico
- large Residual in lsqnonlin
- Prev by Date: Re: Get coordinates from Image?
- Next by Date: please help
- Previous by thread: Re: large Residual in lsqnonlin
- Next by thread: Re: large Residual in lsqnonlin
- Index(es):
Relevant Pages
|