Re: fitting, step

From: Maurice Leutenegger <Maurice.A.Leutenegger_at_email.domain.hidden>
Date: Wed, 7 Jul 2010 19:46:15 -0400
  Mike, Dave, John,

thanks for your suggestions. I tried changing the fit method to subplex, 
and it was able to fit correctly. But lmdif doesn't work at all. So for 
now I will use subplex for my fitting, but let's continue to work on 
getting lmdif to work as well, because there's no reason it shouldn't.

Here are some other things I tried that didn't seem to change anything:

1. My spectrum is in counts (integer). I define the "err" field in 
define_counts using sqrt (counts).
So I tried the following: a.) make counts into a double instead of an 
int; b.) define err as the gehrels variance instead of just sqrt 
(counts), so that there are no fields with variance zero. Neither seemed 
to help.

2. I realized that I was being sloppy in defining "bin_lo" and "bin_hi" 
in define counts. My data file defines them in eV, and I am converting 
to A before doing define_counts. But of course the low bins in energy 
are the high bins in wavelength; and an array that ascends in energy 
descends in wavelength. I hadn't accounted for either of those problems. 
So I swapped bin_lo and bin_hi, and reversed all arrays. No change.
(In fact, I'm surprised that it didn't fail completely with my old bin 
edge definitions. Very robust, John!)

3. I tried using different fit statistics with different fit methods. I 
used "cash" and "chisqr;sigma=gehrels" with both "lmdif" and "subplex". 
Result: both statistics fail for lmdif, both work with subplex.

If I haven't figured it out by tomorrow, I will post a sample code 
demonstrating the problem.



You received this message because you are
subscribed to the isis-users list.
To unsubscribe, send a message to
isis-users-request_at_email.domain.hiddenwith the first line of the message as:
Received on Wed Jul 07 2010 - 19:46:29 EDT

This archive was generated by hypermail 2.2.0 : Tue Nov 13 2012 - 08:45:43 EST