[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gsl] GSL nonlinear least-squares fitting: error calculation
From: |
Brian Gough |
Subject: |
Re: [Bug-gsl] GSL nonlinear least-squares fitting: error calculation |
Date: |
Tue, 18 Mar 2003 22:32:29 +0000 |
Andreas Wolf writes:
> The calculated error is not changing when the data is multiplied by
> a factor. I have made the following changes (main program, line
> 50(?)):
>
> y[i] = 10*(1.0 + 5 * exp (-0.1 * t) + gsl_ran_gaussian(r, 0.1));
> //factor = 10
>
> the output errors of A and b do not change, only the lambda error
> is reduced by the factor. If the calculated errors are absolute
> errors, shouldn't the A and b errors increase by the same factor
> the data is multiplied with? I hope that you can help me with this
> problem.
Thanks for the bug report.
I believe the code is correct, however the documentation probably
needs to make clear that the chisq/degree of freedom should also be
computed.
The error is only reliable when chisq/dof ~=~ 1. In this case the
factor of 10 will increase the chisq/dof to O(100).
best regards,
--
Brian Gough
----------------------------------------------------------------------
Network Theory Ltd Phone: +44 117 3179309 (UK: 0117 3179309)
15 Royal Park Fax: +44 117 9048108 (UK: 0117 9048108)
Bristol BS8 3AL WWW: http://www.network-theory.co.uk/
United Kingdom Email: address@hidden
----------------------------------------------------------------------