[Issues] AssociateList mixes up SIDs

[Kor G. Begeman]

L.S.

The problem was in the sorting and renumbering of AIDs when associating 
an AssociateList with a SourceList which could(!) give unpredictable 
results due to the formulation of the query. Now the query has been 
formulated in such a way that first things are sorted and then 
renumbered which should give the results we expect! This problem has 
been in all versions (< 1.97) of the code and might only occur when you 
have sources distributed as described by Hugo!

Kor

Hugo Buddelmeijer wrote:
> Hi all,
> 
> I noticed some strange behavior in the AssociateList. Under some
> circumstances, the AssociateList seems to store the wrong SIDs with an
> association. It appears to calculate the associations correctly, but
> mixes them up when storing the associations, or something like that.
> 
> The problem is described in the example below. The images referenced to 
> can be found on http://www.astro.rug.nl/~buddel/awe/associatemixup/ .
> The awe scripts used to recreate the problem and the images are attached.
> 
> 
> * Setup *
> 
> I create a master association of three SourceLists, in a two step
> process. The set_single_out_closest_pairs is set to False to preserve
> all associations. The search_distance is sufficiently low to prevent
> unwanted associations (1 arcsecond). The search_area is small enough to
> only have a handfull of sources in each input SourceList. See
> associateSmallAreaShort.py for the script.
> 
> The input SourceLists are from SDSS and UKIDSS. Their SLID and the
> number of sources in the search_area are in order of association:
> SLID   Name               number_of_sources
> 61804  PhotoObjAll (POA)  3
> 155041 Photoz (PZ)        3
> 244531 lasSource (LAS)    3
> 
> Figure allsourcesrealpos.png shows the 9 sources in their real position
> on the sky. Figure allsources.png shows the sources shifted by about 1/3
> arcsecond for enhanced visibility. There are clearly 3 groups. The
> upper-right group contains a LAS singleton which will further be
> ignored. The lower-right group contains a group with one source in each
> SourceList. The upper-left group contains two sources in both POA and
> PZ, and one in LAS. In reality there is only one source there, but it is
> measured twice by the SDSS people, hence it is twice in the POA and PZ
> SourceList.
> 
> 
> * The associations *
> 
> The first association between POA and PZ goes as expected. The 2
> lower-right sources are associated as 1 association and the 4 upper-left
> sources are associated in 2x2=4 ways giving a total of 5 associations.
> See figure assoc100651.png, the colored lines are drawn between the
> associates, the cross represents the search_distance of 1 arcsecond.
> 
> The second association between the first and LAS goes wrong. 3 of the
> previous 5 associate-groups are associated with the correct LAS source,
> but 2 are associated with the incorrect one. The incorrectly associate
> sources seem to be switched. (As expected there also is a LAS singleton,
> resulting in 6 associations.) See figure assoc100661.png.
> 
> 
> * Afterthoughts *
> 
> The resulting associations are way beyond the search_distance. The
> search_distance can be sufficiently decreased so it is small enough to
> not be able to make the correct associations between POA and LAS
> anymore. When this is done, the incorrect associations disappear as
> well. This lead me to believe that the associations are correctly
> determined, but then incorrectly stored.
> 
> The file plotSmallAreaShort.py contains the plotting script. The
> plotting algorithm can be somewhat obscure, mainly because it shifts the
> sources and the association lines by some random amount for improved
> legibility. At the bottom of the script is included a sample awe code to
> plot the associations as simply as possible to show that the strange
> associations are not an artifact of the plotting script but really in
> the AssociateList.
> 
> 
> Hugo
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Issues mailing list
> Issues at astro-wise.org
> http://listman.astro-wise.org/mailman/listinfo/issues

-- 

Dr. K.G. Begeman
OmegaCEN
Kapteyn Institute
University of Groningen
Postbus 800                         NL-9700 AV Groningen
Landleven 12                        NL-9747 AD Groningen
The Netherlands
Telephone                           +31-(0)50-3632454/4073
Telefax                             +31-(0)50-3636100
e-Mail                              kgb at astro.rug.nl
WWW                                 http://www.astro.rug.nl/~kgb