[scilab-Users] covariance matrix and leastsq

[Antoine Monmayrant]

Cool, thanks a lot.
I'll give at try today and see how it fits my needs!
(pun intended)

Antoine

On 04/28/2010 08:30 AM, Eric Dubois wrote:
> Hi.
>
> In my package Grocer (available trough Atoms or on my web site :
> http://dubois.ensae.net/grocer.html), wirtten with Emmanuel Michaux, you
> will find among other things a function ols that provide standard errors of
> coefficients as well as the covaraince matrix of the coefficients.
>
> Eric.
>    

> 2010/4/26 Antoine Monmayrant<antoine.monmayrant at laas.fr>
>
>    
>> Hi everyone,
>>
>> I am using leastsq to fit some experimental data with different kind of
>> model functions.
>> In order to properly analyse the result of these fits, I would like to get
>> an estimate of the standard error on each parameter (or an interval of
>> confidence at 95%).
>> So far, the only way I found to retrieve the interval of confidence is to
>> do bootstrapping but it requires repeating the fitting procedure ~ 100
>> times.
>> I am looking for a less computer-intensive solution.
>> As far as I understood, leastsq is using minpack to implement the
>> Levenberg-Marquardt algorithm.
>> Is there a way to get access to the covariance matrix with leastsq, optim
>> or any other scilab routine?
>> This would be way faster and easier than bootstrapping.
>>
>> Thanks in advance for your help,
>>
>> Antoine
>>
>>
>>      
>