| Re: Fitting several Gaussians [message #2850 is a reply to message #2773] |
Fri, 23 September 1994 17:00  |
agraps
Messages: 35
Registered: September 1994
|
Member |
|
|
gotwols@strdev.jhuapl.edu (Bruce Gotwols) writes:
> I am looking for a fitting routine that will do a nonlinear least
> squares fit of the sum of N Gaussian functons to measured data. I am aware
> of the routine in the IDL Users library that fits a single Gaussian plus a
> polynomial, but in my case I want to fit 3 Gaussian functions (or even
> better yet, N). If anyone knows of the whereabouts of such a function I would
> appreciate hearing about it, other wise I will have to write it myself.
> Thanks in advance, bruce
> --
> Bruce L. Gotwols
> Johns Hopkins University, Applied Physics Lab., Laurel MD 20723
> Internet: gotwols@tesla.jhuapl.edu (128.244.147.15)
> SPAN: APLSP::STR::GOTWOLS
You can use the same routine CURVEFIT in the IDL library. What will be
different will be your FUNCT.PRO function. It should be straight-forward,
maybe just a few lines of code.
In the IDL library, FUNCT.PRO has the Gaussian function defined as
(I've left off the polynomial):
F = A(0)*EXP(-Z^2/2)
What you could do instead is pass in your "A" estimate array defining the
1st estimates for your number of Gaussians:
PGauss = N_ELEMENTS(A) -1
Then, initialize the argument, using the N from the length of your
indep x array:
N=N_ELEMENTS(X)
F = DBLARR(N) + A(0)
Then create your F, term by term:
FOR K=1, PGauss DO F = F + A(K) * EXP(-Z^2/2)
Another way to do this problem is make it linear:
take the log of it, and do a linear least-squares fit. (but
if you have polynomial terms, then this approach won't work.
(If someone else has answered this, sorry my response is late.
I'm experiencing a delay in the newsfeed at my Internet
provider site.)
Amara
************************************************************ ********
Amara Graps
Computational Physicist
Intergalactic Reality Boycott the WireTap Chip!
agraps@netcom.com
************************************************************ ********
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
