RE: Soliciting criticism

["Creager, Robert S" <CreagRS@LOUISVILLE.STORTEK.COM>]

Michael,

Thanks for the prod.  In setting up for the Welch-Stetson statistics, I was
verifying the weighted mean, and it was occasionally very different from the
normal mean.  A little digging brought up a problem.  SPS2X doesn't spew out
wildly inaccurate magnitudes and error estimates, but sextractor can.  I was
not filtering out sextractor's far out values in one SQL statement (but I
was in another), with the result of including magnitudes with values of >
100 (pretty faint), and an error magnitude of 99 being included.  The
weighted mean showed this error very clearly.  I will be re-generating the
sextractor graph later tonight.  I had SPS filtering out large (> 1) error
estimate stars, so those graphs are more constant, but still suffer from the
same problem.  These values appear about 10% of the time in sextractor
(although sextractor generates more stars than SPS, so the total usable star
count is nearly a wash).

Chris,

Nice pick up on SPS and the constant PSF.  The scatter makes more sense to
me now.  SPS does calculate a constant, two dimensional (I believe) PSF from
stars it picks.  With your prod, I now understand this is what I was seeing
in the SPS run logs.  I may have a go at the central portion of the image.
I'll wack out some software which clips the outer edges of the image, and
re-run SPS and see what comes out.  What if the image was broken into 9 (or
16) sub images, and each reduced separately?  Painful, but possible.  Anyone
interested in seeing this done?

Now, are there any packages which use non-constant two dimensional PSF
models?  Andrew, is your software ready for testing purposes?  I have been
admiring your order-of-magnitude-better-than-mine-graphs for a while now :-)
You're running Pascal in DOS, right?  I'll be willing to look at what it
takes to port to Linux if you're interested.

I'll see if I can answer my own question, after I send this of course...

Thanks for the input,
Rob


> -----Original Message-----
> 
>   It puzzles me that the two graphs are so different.  In one case,
> the I-band data appear to have smaller scatter at a given magnitude;
> in the other, the V-band data.  I would guess that there are 
> significant
> tweaks you can make to the parameters which control the software.
> In theory, one might expect the same results from two programs running
> on the same data, no?  
> 
>   I'm afraid I'm not familiar with either program, so I can't
> suggest which parameters you might change, and how much.  Sorry.
> 
>   Here are my suggestions:
> 
>      - try to figure out why one piece of software does better on
>              the V-band, and the other on the I-band.
>              Perhaps one camera has sharper images, and one
>              program's parameters turn out to work better on sharp
>              images?
> 
>      - pick one set of results, both V and I.  For stars with >= 20
>              measurements, calculate the Welch-Stetson statistic.
>              Select those stars with very large values of this
>              statistic, and make some light curves.  Some of the
>              chosen stars will be contaminated by single bad 
>              measurements -- ignore them.  Try to come up with 
>              a way to pick out the stars which really do vary more
>              than they ought to.  Then look:
> 
>                   trends versus position on chip?
>                   trends versus color?
>                   any matches to catalogs of known variable stars
>                           in the area?
> 
>   That's what I'd do.
> 
>                                                 Michael 
>