Levenberg–Marquardt doesn't do a good job fitting

[Pith]

While I'm sure this has more to do with how I'm using it I was wondering if you guys could give my program a look over.

The program should be fitting the data to a sigmoidal curve, but doesn't manage to and instead assumes all the data is apart of the upper asymptote I believe.

I just copy the data from TestFile and open up curvefitting and hit the "Paste" button, which pastes the data into the indicator. It then feeds into the program. I get an error with the curve fitting program, but if I hit continue it just gives me a bad fit. Wondering if anyone can help me get this to work.

Thanks,

-k

[johnsold · Accepted solution]

Did you look at what you initial guesses produce when plotted over x? The curve goes the wrong way (from top left to bottom right) and thus does not fit the data.

Either the formula should have y0 - ((.... or the initial guess for a should be negative.

Why are you using the Formula Express VI to calculate something that the Lev-Mar VI already does for you? Just plot the "best nonlinear fit" output against the input x.  I am not sure why the Express VI gives a different result.

Lynn

[Pith]

Woo! thanks, both things worked.

[johnsold]

Any time you are doing significant analysis of some data it is a good idea to plot and look at both the data and the analysis steps.  This can often help you spot errors or just "that does not look right" so you can check what is going on.

When I looked at your fit and saw a straight line, it was obvious that something was seriously wrong.  Then I plotted the results of using the initial guesses against x. This is what clued me to the sign error. Fitting is often quite sensitive to the initial estimates, so checking those against the original data can be very useful.

Lynn