CASB Report August,
2001
· Enhanced the hand-classification software to generate a training set that is measurably more representative of the real objects being classified.
· Hosted several tours of the plate vault and scanning facility including flight directors from JSC.
· CASB sends best wishes to members of its staff who are departing or transitioning.
o Gretchen Greene will increase fraction of time matrixed to the Independent Detector Testing Laboratory.
o Fabrizia Guglielmetti will resume graduate studies and perform research on the ROSAT All-Sky Survey at the International Max-Plank Research School for Astrophysics in September.
o Vicki Laidler is in transition to the ESS Science Software Group.
o Jane Morrison is assuming a position at the Steward Observatory, University of Arizona, in October.
o David Wolfe has joined the ESS Data Systems Team and is matrixed with CASB.
·
Ricky Smart
(OATo) and Tiziana Franconieri (student in Multimedia Arts and Performance at
Torino University) are visiting. They
have prepared a demo CD-ROM presenting an introduction to GSC-II that is
suitable for a general audience. CASB
staff are providing additional material for the CD-ROM.
·
McLean and Greene continued supporting SDSS planning
activities.
· Reviewed ten Guide Star Problem Reports (GSPRs) this period. All the GSPRs were fine lock failures. Filed a Bad GS Alert for each of the GSs.
· Work is continuing on the Bright Object Tool in the VTT with the next release scheduled in the December timeframe.
· No activity
· Continued scanning POSSI-O plates until the latest shipment of UKST-IR plates were available. Scanned the UKST-IR plates the rest of the period.
·
Continued
compressing scans of the POSSII-N survey.
·
Continued
compressing scans of the POSSI-O survey.
· Continued planning for porting the GetImage software from VMS to NT.
· Began migration of DSS1 from CDROM to online network storage..
· Worked on three archive help desk requests for help/information.
· Continued low-level rationalization of MOD archive.
·
The GSCII database
server, COMPASS, has been upgraded to Windows 2000 as well as a new version of
Objectivity (6.1). New hard disk
storage (1.6TB RAID 5) was also added.
·
Planned a new database
file structure and started its implementation. This should allow maximum
efficiency of the new disk storage. The
file moves are going well and should be completed very shortly.
·
(OATo) Developed tools
for plotting all-sky high-resolution maps of the GSC2 objects. These can be useful for the detection of
plates affected by various kinds of problems.
·
(OATo) Performed
comparisons between GSC2 photometry and deep CCD catalogs. Typical errors of
0.1-0.3 mag are found.
·
(OATo) Comparisons
between GSC2.2 and UCAC/2MASS give indications that the decision to select
objects nearest to the plate center has reduced the effect of magnitude terms
in producing systematic errors.
· Working on understanding the systematic errors in GSC 2 positions, which are a function of magnitude and distance from the center of the plate. Using the SPM to quantify the effect.
· Modified the XSPAM hand classification tool, hcl_mark, to select objects for hand classification based on their distance from the domain occupied by existing training set objects. For the first time, we have an approach to building training sets that is directed toward a specific and measurable goal. Adding training set objects in this way will extend the domain of the training set to occupy more of the space occupied by real objects – thus producing a more representative training set. This approach also permits the human classifier to better concentrate on the actual task of visual classification, by freeing her or him from most of the decision making over which objects to examine.
· Created a new training set construction task, VERTRAIN, as part of the VERCLASS package. When run over an OOP file that contains hand-classified objects, it produces a FITS file containing a training set with ranked features and identifying information. This FITS file can be used by OC1 to generate a decision tree. This eliminates the need to run classify in a special "training" mode, creating a temporary OOP file for each plate with ranked values overwritten in place of raw values. Although this task (still in development) requires a few more capabilities in order to be practically useful, it is the last missing link in the suite of software tools required to manage training sets and decision trees in a sensible way.
· Continued modifications to VERCLASS, the versatile classification task presently in development. The voting procedure has been modified to use the object’s distance from the training set domain for a given tree in weighting its vote. Experiments comparing the results of weighted voting from all-tree voting indicate that that this change in voting procedure, while still affecting only 1-3% of the objects, produces changed classifications that are correct in 70-80% of the cases: a clear improvement.
· Wrote an XSPAM utility, distance_distribution.pro, to assist in tuning the weighting function, which is an exponential of the form exp(-D/scale). It produces plots showing the distribution of training set domain distances, and up to two possible weighting functions, for each of the five trees. Experiments indicate that a scale factor of 0.017 is reasonable for most plates; 25um scans are the notable exception. This utility will need to be rerun, and the scaling factor retuned, whenever a new set of decision trees is introduced.
· Implemented an algorithm to define the classification unreliability in VERCLASS. This value is defined to range from 0 to 1, where 0 is reliable and 1 is unreliable. Presently, the distance from the training set domain, and the weighted number of voters, are combined to produce this value. A document that discusses the unreliability in detail has been drafted. The contribution of the number of voters requires some further analysis.
· A small bug due to rounding error was detected in VERCLASS. Very occasionally, the ranked value (defined in the range {0,1}) of a feature for a given object will be computed to be greater than 1 by about 1e-6. This occurs when the ranked value is averaged across neighboring regions. A workaround was implemented in the tree domain code, which was otherwise excluding objects whose ranked integrated density was greater than one.
CALIBRATIONS: Photometry
(GSPC-II and DSS-I and -II)
· After reducing the data from La Palma Run 5 (June 23 to July 1) we found nights 1,3,7 and 9 to be photometric quality. The data included 15 POSS-I, 16 POSS-II, and 4 southern sequences.
· (OATo) White Dwarfs project: processed six new fields in the summer sky and got a new list of candidates. Observed one of these targets on August 14 at TNG (La Palma); the reduction is in progress.
· (OATo) Preparing a paper about the rare C dwarf star that was serendipitously discovered during the WD search last spring.
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
|
POSS-II
J |
897 |
902 |
0 |
902 |
100 |
S4 |
1 |
|
POSS-II
F |
897 |
910 |
0 |
910 |
100 |
S4 |
1 |
|
POSS-II
N |
897 |
880 |
0 |
880 |
98 |
S4 |
1 |
|
AAO-SES |
606 |
614 |
0 |
614 |
100 |
S4 |
1 |
|
PPARC-ER |
288 |
288 |
0 |
288 |
100 |
|
1 |
|
AAO-SR |
118 |
118 |
0 |
118 |
100 |
S2 |
- |
|
UKST-SR
|
112 |
50 |
0 |
50 |
45 |
S5 |
3 |
|
UKST-IR |
894 |
697 |
36 |
733 |
82 |
S6 |
2 |
|
SERC-J/EJ |
447 |
443 |
0 |
443 |
99 |
S1 |
2 |
|
POSS-I
E |
106 |
106 |
0 |
106 |
100 |
S3 |
3 |
|
POSS-I
O |
936 |
354 |
5 |
359 |
38 |
|
3 |
|
Rescans |
|
34 |
0 |
34 |
|
|
|
|
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-II
J |
897 |
895 |
0 |
897 |
100 |
|
2 |
POSS-II
F |
897 |
897 |
0 |
897 |
100 |
|
1 |
POSS-II
N |
897 |
681 |
39 |
720 |
80 |
|
3 |
AAO
SES |
606 |
606 |
0 |
606 |
100 |
|
1 |
PPARC
ER |
288 |
288 |
0 |
288 |
100 |
|
1 |
AAO-SR |
118 |
118 |
0 |
118 |
100 |
|
3 |
UKST-IR |
894 |
697 |
0 |
697 |
78 |
|
3 |
POSS-I
O |
936 |
20 |
28 |
48 |
5 |
|
3 |
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-II
J |
897 |
902 |
0 |
902 |
100 |
|
4 |
POSS-II
F |
897 |
900 |
0 |
900 |
100 |
|
3 |
POSS-II
N |
897 |
833 |
0 |
833 |
93 |
|
4 |
AAO-SES |
606 |
613 |
0 |
613 |
100 |
|
3 |
PPARC-ER |
288 |
288 |
0 |
288 |
100 |
|
3 |
AAO-SR |
109 |
116 |
0 |
116 |
100 |
|
4 |
UKST-IR |
894 |
697 |
0 |
697 |
78 |
|
5 |
SERC-J/EJ |
894 |
892 |
0 |
892 |
100 |
|
|
SERC-J/REPAIR |
48 |
48 |
0 |
48 |
100 |
|
|
SERC-J/15mu |
449 |
449 |
0 |
449 |
100 |
|
5 |
POSS-I
E |
933 |
928 |
0 |
928 |
99 |
|
2 |
POSS-I
REPAIR |
198 |
198 |
0 |
198 |
100 |
|
2 |
POSS-I
O |
936 |
20 |
50 |
70 |
7 |
|
5 |
POSS-QV |
613 |
613 |
0 |
613 |
100 |
|
- |
SERC-QV |
94 |
94 |
0 |
94 |
100 |
|
- |
GPO/BB |
88 |
88 |
0 |
88 |
100 |
|
- |
|
|
|
|
|
|
|
|
GSC-II PLATE PROCESSING
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-I
E |
936 |
931 |
0 |
931 |
99 |
|
3 |
POSS-QV |
613 |
2 |
0 |
2 |
0 |
|
3 |
POSS-II
J |
897 |
897 |
0 |
897 |
100 |
|
1 |
POSS-II
F |
897 |
897 |
0 |
897 |
100 |
|
1 |
POSS-II
N |
897 |
232 |
76 |
308 |
34 |
|
4 |
SERC-J/EJ |
894 |
829 |
0 |
829 |
93 |
|
1 |
South-F |
894 |
894 |
0 |
894 |
100 |
|
1 |
UKST-IR |
894 |
168 |
66 |
234 |
26 |
|
4 |
South-SR |
116 |
116 |
0 |
116 |
100 |
|
2 |
South-QV |
94 |
94 |
0 |
94 |
100 |
|
2 |
GSC-II COMPASS DATABASE LOAD (no.objects = 998,402,801)
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-I
E |
936 |
23 |
0 |
23 |
2 |
|
3 |
POSS-QV |
613 |
2 |
0 |
2 |
0 |
|
3 |
POSS-II
J |
897 |
897 |
0 |
897 |
100 |
|
1 |
POSS-II
F |
897 |
899 |
0 |
899 |
100 |
|
1 |
POSS-II
N |
897 |
27 |
0 |
27 |
3 |
|
4 |
SERC-J |
894 |
709 |
0 |
709 |
79 |
|
1 |
South-F |
894 |
742 |
0 |
742 |
83 |
|
1 |
UKST-IR |
894 |
2 |
0 |
2 |
0 |
|
4 |
South-SR |
116 |
116 |
0 |
116 |
100 |
|
2 |
South-QV |
94 |
90 |
0 |
90 |
96 |
|
2 |
GSPC-II OBSERVATIONS
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-I |
584 |
588 |
15 |
603 |
100 |
P1 |
1 |
POSS-II |
314 |
255 |
16 |
271 |
86 |
|
1 |
SERC |
894 |
885 |
4 |
889 |
99 |
|
1 |
GSPC-II REDUCTIONS
Survey |
Goal |
Previous Month |
Current Month |
Current Total |
% done |
Notes
|
Priority |
POSS-I |
584 |
588 |
15 |
603 |
100 |
P1 |
1 |
POSS-II |
314 |
255 |
16 |
271 |
86 |
|
1 |
SERC |
894 |
902 |
4 |
905 |
100 |
|
1 |