| Author |
 Topic  |
|
|
randoob
USA
2 Posts |
 Posted - 04/29/2003 : 3:33:35 PM
|
What is the definition of the "Error" and "Dependency" values that you see alongside the optimal parameter values found by a nonlinear curve fit? The Numerical Methods book, referenced by Edgar.Kaiser a few threads back, presents both Levenberg-Marquardt and simplex methods used by the NLSF for curve fitting. If the book explains what the "Error" and "Dependency" are, it is in language I do not recognize.
If your model is y = a*f(x) + b, and the curve fit reports that a=1 with an "Error" of +/-100000, does that mean that a could just as well be negative or even zero?
Edited by - randoob on 04/29/2003 3:35:40 PM |
|
|
edgar.kaiser
Switzerland
29 Posts |
Posted - 04/30/2003 : 04:18:54 AM
|
You can find an explanation in the Origin help using the index string:
Nonlinear least squares fitter, parameters:dependency
Your assumption appears to be correct.
Regards,
Edgar |
 |
|
|
easwar
USA
1965 Posts |
Posted - 04/30/2003 : 09:36:42 AM
|
Hi Randoob,
The error value quoted is the standard error on the parameter value. The dependency, which is computed from the variance-covariance matrix, typically indicates the significance of the parameter in your model. For example, if some dependency values are close to 1, this could mean that there is mutual dependency between those parameters - in other words, the function is over-parametrized.
Now, if you are getting such a large error of +/- 1000000 on a parameter value of 1, this most probably is indicating that the fit is not very good - could be due to various reasons such as wrong choice of model for the data, not enough data points etc. Probably the chi-sqr value you get is also very large?
If you need further help, please contact your tech support office with your data, and fitting function etc.
Also, take a look at this article:
http://www.tipmagazine.com/tip/INPHFA/vol-9/iss-2/p24.html
Easwar
OriginLab.
|
|
|
| |
Topic |
|
|
|
"Error" and "Dependency" in Nonlinear Curve Fit