Drizzlepac/HAP: test new alignment parameters in SVM regression test run

[stscijgbot-hstdp]

Issue HLA-1422 was created on JIRA by Steve Goldman:

I've asked Matt to generate a run of SVM regression tests to see the effects of the new alignment parameters. I (along with instrument scientists) will see how these changes affect the alignment and catalogs.

[stscijgbot-hstdp]

Comment by Steve Goldman on JIRA:

I've been testing the impact of the new alignment configuration parameters along with Jennifer Mack. Attached is a csv list of important header keywords in the total SVM products the matt ran, both for the baseline directory (photutils build), and our alignment configuration parameters. 

data location:

alignment configuration paramters: /ifs/archive/dev/processing/hla/home/mburger/singlevisits/results_2025-01-29 

photutils baseline: /ifs/archive/dev/processing/hla/home/mburger/singlevisits/results_2025-02-03```
You can see from the first few lines of the table that we are seeing what we expect. The first row shows 4 nmatches and a SVM_GaiaEDR3 WCS solution. The new alignment configuration parameter change requires a higher nmatch threshold of at least 12 for absolute fitting. We see 12 nmatches and an IMG_GaiaEDR3 WCS solution. The later solution is from pipeline processing, the former, from SVM processing. The second row shows a similar pattern, and so on.

 

I also see slight changes in the RMS_RA and RMS_DEC when the same WCS solution is achieved, and some cases where the same WCS solution is achieved, but with a different number of nmatches. 

This can be explained by the fact that the alignment configuration parameter branch did not include the photutils changes. 

In conclusion, the alignment parameter changes look good to me!

[stscijgbot-hstdp]

Comment by Steve Goldman on JIRA:

We did another test run of the photutils+alignment configuration parameters (2/6 test run). I compared our photutils baseline (2/3 test run) with the new test run and attached the resulting wcs related header keywords as csv files. 

While it seems like we have more matches for the programs with few matches, it seems as though we generally have high rms values. 

!Figure_2.png!
Here is a scatterplot of the difference in the RMS_RA for all drizzled total SVM products that Matt ran, for the two runs. I have removed programs that have the same alignment solution in both runs (of 1938 total svms). As an aside, this include the total drz svm files for the UV products that also have drc files. In the figure, the version of the code with the updated alignment configuration parameters has higher RMS_RA values than before. In many of these cases, we no longer achieve a GAIA alignment, and instead an alignment to GSC242. Below I also show a similar plot that shows the difference in the number of matches (nmatches) between sources and catalog sources. 

!nmatches.png!