David Site Admin
Joined: 25 Nov 2006 Posts: 1964 Location: Chapel Hill, NC

Posted: Sat Mar 24, 2018 11:24 am Post subject: 


The issue might be that the range is too large for the standard least squares fit. In DataGraph, this is fixed by allowing you to fit it logarithmically, but in DataTank this option is not available.
Here is one way to explain it. If the data is approximately 0.1*exp(x) + 10*exp(5x) on for example [0,6] the values are very small on the portion where x>2. So if f(x) = a*exp(bx)+c*exp(dx), instead of minimizing the distance f(x)y(x) you minimize log(f(x))log(y(x)) = log(f(x)/y(x)). That is esstentially minimizing the relative error instead of the absolute error.
If this is a DataGraph question, that should fix it. If this is a DataTank question the workaround is to map the data using log and then add log to the fit function.
David 
