Changeset 25755 for trunk/psphot/doc/notes.txt

Timestamp:

Oct 2, 2009, 3:12:47 PM (17 years ago)

Author:

eugene

Message:

check in changes from gene@development branch : extensive changes to moments calculation, psf model generation, aperture residuals; improvements to extended source analysis

File:

• trunk/psphot/doc/notes.txt (modified) (1 diff)

trunk/psphot/doc/notes.txt

@@ +1 @@
+
+2009.09.26
+
+  Remaining PSPHOT Issues:
+
+  o soften errors a bit for bright stars?  (essentially allow a
+    systematic error on the flat of some percentage).  this may help
+    the very saturated stars (peaks > 10k) which seem to be
+    excessively driven by the cores.
+    ** did not seem to be a big success --> makes things worse for f =
+    0.01, about the same for f = 0.005
+
+  o clean up and remove test bits of code.
+
+  o clean up psphot.config files : drop unused, fix some conflicting
+    names, reduce camera-specific options, make S/N limits
+    consistent...
+
+  * PSF Eval & ChiSq test??
+
+  * EXT fail fit?  lack of iterations?
+
+  * extended source output
+
+  * still some concerns with the source size analysis: the parameters
+    space seems fine (Mxx, Myy, nSigma), but we need to define more
+    sophisticated boundaries (perhaps include dMag or Mag).  
+
+  * some optimization issues:         
+    * too many re-calculations of all magnitude / aperture data
+    * is FluxScale really faster?  do we need a finer mapping? (5x5
+      clearly yields too much scatter)
+    * is psphotFitSourcesLinear.c (build matrix) super-slow when optimized?
+
+  * very saturated stars are somewhat segmented.
+
+  * footprint coverage in dense simtest image 005.001?
+
+  * footprint S/N limit vs peaks S/N limit?
+
+  * warning about burntool table?
+
+  * is the residual map failing to be masked at the radius?
+
+  * PSF 2D convergence : how to choose between marginally different
+    options?  (note that the GPC1 images may go to 2x2 instead of 3x3)
+
+  * -visual needs more fine-grained control
+
+  * visualization graphs should reset the sections in general.
+
+  * some excess verbosity (my log level is 9, lower?)
+
+  ----- 
+
+  Apparently I'm getting bad psf vs ap mags because of the difference
+  between linear and non-linear fits.  I think this is due to the
+  weighting schemes:
+
+  psphotChoosePSF & psphotBlendFit depend on POISSON.ERRORS.PHOT.LMM,
+  which is set to TRUE => use Poisson errors
+
+  psphotFitSourcesLinear depends on CONSTANT_PHOTOMETRIC_WEIGHTS,
+  which is set to TRUE => do not use Poisson errors.
+
+  nothing seems to use the value POISSON.ERRORS.PHOT.LIN
+
+  ** Poisson errors seeem to be working much better than Constant
+     errors (as judged by consistency between PSF and APERTURE
+     magnitdues).  Not sure I understand this.  It could be that the
+     implementation of Constant errors in the linear fitting analysis
+     is busted.  
+
+  ** Also, I was getting some additional scatter from the FluxScale
+     (lookup-table for Io -> mags).  
+
+     * surprisingly, it is not clear that the FluxScale method is
+       faster than the more accurate integration technique...
+
+2009.09.25
+
+  Notes on pmSourceMagnitudes:
+
+  * in psphotChoosePSF, in pmPSFtry: raw PSF & raw AP
+  * in psphotChoosePSF, psphotMakeGrowthCurve: uses direct all to pmSourcePhotometryAper
+  * in psphotSourceSize, in psphotSourceSizePSF, raw PSF & raw AP (but uses moments->Sum) for PSF stars
+    NOTE: on the initial call, the PSF stars may have had multiple models tried and the psfMag
+    may not be the value for the chosen model.  
+  * in psphotSourceSize, in psphotSourceClassRegion, raw PSF & raw AP for all sources for which size is not yet measured.
+  * in psphotApResid: raw PSF & raw AP for STARS only
+  * in psphotMagnitudes; corrected PSF & AP mags
 
 2009.01.24