How to specify required precision for NonLinearCurveFit

[Alok Damle]

Hi All,

I am using the Non Linear Fit method. The method takes all the argument accept the stopping crieteria. Only crieteria it takes is the max iteration. I want the Algorithm to iterate till the Mean Square error of 1E-08 is reached. However it seems the exit crieteria is hardcoded to be 0.001 as the value of the minimum Marit function deviation.

I am working with very small numbers here and hence require quite higher level of precision. 

Can anybody tell me the workaround to get higher level presicion (like whether scaling would help? if so then what is the exact procedure? i tried my way of it however no luck). I have another software (which is a freeware) using which when i do this fit, the residual is very minimum (less than 1E-10, very much in my acceptable range) but when i use NI library (for which i have paid a realy hefty amount) it stops pretty early whithout allowing me to specify the required precision. 

I know updates to the library would not be that early however any idea on any kind workaround is appreciated.

And I strongly suggest to please add this facility in the whole of CurveFit library in your future updates. Any professional library must allow the user to decide what level of precision is good enough!

Regards,

Alok

[Alok Damle · Accepted solution]

Its been about a week since i posted this issue but not a single response from NI people. SAD!

However after a long serching over internet found following link

http://www.datamaster2003.com/

For all those who are stuck with NI's CurveFit package and its limitations, here is the simplest way to use any kind of curvefitting. following are the features:

1. Single library for Linear as well as Non-Linear fitting

2. User can define threshold and weights to parameters

3. Also user can define the function by string expression. Hence no issue of changing the definition of the Function to try out new possibilities of functions.

I was very much moved with this activeX control which is a FREEWARE. May be NI people should learn how a library should be when they charge astronomical values for Measurement Studio and give you implementations where user does not have flexibililty to even specify the threshold. I guess majority of implementations from the MS is better suited for acedemic research only. 

For commercial usage this FREEWARE is far more superior than Measurement Studio.

Regards,

Alok Damle.