> Malcolm Currie is looking into it, (see an email he sent below), so I guess
> it is best to wait to see what the explanation was.
Regressing to use the exclude most-likely candidate fix from last
December, still does not generate silly fits. It is only the masking
that's affected in that fewer pixels are masked with the bug. I've not
been able to reproduce the problem with the development code and some
older versions of MFITTREND subroutines. I suppose I could download
the lehuakona source, rebuild everything, and debug over again. That
would start with the known instance, and might reveal the extent of
the damage.
> One critical question for NGLS is how much does the occurrence of the bad
> fitting bug depend on the input .sdf and the command line parameters of
> mfittrend. This is probably easier to answer by looking at the code
> differences in the KAPPA versions rather than repeated testing on lots of
> cubes.
It doesn't appear so. The effect of some bugs isn't predictable and
surprising results can ensue. Sometimes coming back to code, I'll find
a bug and wonder how the code ever worked. There are compiler
differences too. Given there was a fix made somewhere yet it's only
been noticed for the second time after 9 months suggests that it's is a
restrictive set of parameter values, and my gut feeling (based upon the
evidence) is that method=global is the key determinant. Now perhaps
some data are unaffected, but if we can automate the re-reductions we
might as well address all the processed cubes.
As a compromise, could Jamie please try further tests with the known
problem dataset, using all the MFITTREND parameter settings invoked
during the NGLS reductions. Are there any other problem fits other than
when method=global? I can't do this because I don't know how the NGLS
data have been reduced.
I think better use of my time would be to help address how to re-reduce
Christine's data. First can the scale of the problem be assessed by
running HISLIST on the reduced data and search for those NDFs that used
lehuakona (KAPPA 1.9-11) and mfittrend method=global? As Frossie says,
with ORAC-DR it should be possible to re-reduce data quickly, unless
there was a significant manual interaction.
While trying to find more about NGLS, I came across the Wilson et al.
paper on molecular gas in four Virgo spirals. It dismayed me to see
that the software (CUPID and KAPPA should be capitals BTW), just had
footnotes rather than proper citations. Programmers have careers to
worry about too. In future papers please consider this. For CUPID
there's the Berry, D.S. et al., 2007, ASP Conf Series 376, Astronomical
Data Analysis Software and Systems XVI, 425. For KAPPA, I suggest a URL
to SUN/95 pointing to JAC's copy.
Malcolm
|