From: kingsley at S. <kg...@sy...> - 2008-09-28 00:46:33
|
Knut Now I see the "spin box". It is only present for Gauss, Lorentz and polynomial functions. I am actually using GaussAmp (sorry, I was not very precise about that), and it is not available there!!! For some reason the gauss fit does not work well on my data, but gaussAmp does. It would be nice to have that spinbox on the GaussAmp function also! Knut Franke wrote: >> When I make my own "multi-peak-fit" function I do not see a "spin-box" >> that you refer to. > Well, you can use this feature together with custom initial guesses if you use > the built-in Gauss function. If you choose "Built-in" from the "Category" > list, and then "Gauss" from the list of functions beside it, said spin-box > should appear. See the attached screenshot. > >> On some of my data, the best fit occurs with a very large negative >> offset of the y axis. I tried to prevent this by removing the y0 terms >> from the gaussian equation, but that does not seem to solve the problem. >> I am concerned that as my experimental data cannot actually be negative, >> it seems unreasonable to fit it using curves which are more negative >> than the data is positive! Is there a way to constrain the component >> gaussians to prevent them going into negative space? > > Currently not; and a clean solution for constraints on fit parameters will > likely require some consideration. Also, I'm not sure in how far a result > lying on a (essentially arbitrary) constraint border in parameter space is > actually meaningful. I can prevent the component curves of the fit being negative by removing y0 from the equations - or setting it as constant to zero iff i unset the "built in function" tick-box after initially selecting the built in function required. If the fit wizard remains set to "built in function", the ability to declare y0 constant is not available. If different starting guesses don't help, then maybe > reconsidering the curve you're fitting may be a good idea. Possibly the peaks > aren't actually gaussian, or there's some background to be taken into > account. You are correct. There is some concern whether my data is actually gaussian. A co-worker using PLOT on macintosh recently suggested I should use skewed gaussian populations. I have made some user functions by adding a skew term to the GaussAmp built-in function in scidavis and they are better for some cases. Thanks - I greatly appreciate your very useful help and comments. kingsley |