Re: Problem with gaussfit in IDL/v5.5 [message #28034] |
Fri, 16 November 2001 12:41 |
Craig Markwardt
Messages: 1869 Registered: November 1996
|
Senior Member |
|
|
thompson@orpheus.nascom.nasa.gov (William Thompson) writes:
> We found a problem with the procedure gaussfit which is supplied with IDL/v5.5.
> It appears that somebody at RSI tried to improve the algorithm for generating
> the initial guess, but didn't quite complete the job. I've informed RSI, and
> decided to also inform the newsgroup. The simple fix is to change the line
...
Hi Bill--
That's an interesting report. If you are interested in a different
approach to estimating the parameters of peaks, then you might be
interested in MPFITPEAK, which is based on the MPFIT fitting routines.
I was having real problems with noisy data, that didn't work very well
with GAUSSFIT. I found that a better approach is to compute the
*area* above and below the mean, and that gives a much better handle
on the position and width of the peak. It is very robust, even for
very broad and noisy peaks.
However, if this is important to you:
> if nt gt 5 then a = [a, c[2]]
[ indicating you need the quadratic approximation to the background
before you begin fitting] then MPFITPEAK may not be for you. It
doesn't pre-fit the background level, instead assuming it is constant.
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@cow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|