| Author |
 Topic  |
|
|
Dimitris2
USA
Posts |
 Posted - 04/25/2004 : 11:33:58 PM
|
Hi
I have been trying to fit a user defined equation on my data, that has 5 unknowns (and over 600 datapoints so that should be ok). The fit comes out REALLY good when it is plotted out, the actual and model curves almost completely superimposed. However, when I calculalate the parameters worksheet, the error is VERY large (3-4 times the value of the parameter). I know that it cannot be that large, because the starting estimates of the parameters I used were much closer to the final estimates than the errors are, and yet the curve was way off...
any suggestions? |
|
|
easwar
USA
1964 Posts |
Posted - 04/26/2004 : 09:51:18 AM
|
Hello,
One possibility is that your function is over-parametrized. Even though there are only 5 parameters, still one or more of the parameters may not really be needed, and the model could be simplified. If the function is over-parametrized, then one could end up with very large error values for some of the parameters at least.
If the above is not the case, please send your OPJ to tech support so that we can take a look at what else could be the reason.
Easwar
OriginLab
|
 |
|
|
Dimitris2
USA
Posts |
Posted - 04/26/2004 : 10:44:35 AM
|
I discovered what's happening. I was using weighing (direct). If I take it off and do 1 iteration, the error drops from 3xe7 to 17!!!!
cheers though,
Dimitris. |
|
|
| |
Topic |
|
|
|
NLCF problem with errors