Investigate differences between user and pipeline match algorithm in skymatch

[stscijgbot-jp]

Issue JP-3843 was created on JIRA by Ned Molter:

Internal pipeline user Nicolas Flagey ran the skymatch step on the data mentioned above using the "match" algorithm, and compared it to their custom implementation of the same algorithm, which is in the attached Jupyter notebook.  Although the two are supposedly doing the same thing, the Jupyter notebook version seemed to outperform the pipeline version, with the pipeline version showing much larger mismatches in the sky level between exposures than the notebook version.  This remained the case even after persistence and 1/F noise were handled prior to skymatch; the cleaned version of the data is what is in the box folder.  This also remained the case playing with all the various input parameters to skymatch.

It should be investigated what is going on here.

• Are they indeed the same algorithm?
• Can this dataset be processed successfully using the existing pipeline version by setting parameters in a way that Nicolas didn't already try?
• Is there a bug in the pipeline version?
• If not a bug, should improvements be made to the pipeline version?

[emolter]

@mcara What is your take on this?

[stscijgbot-jp]

Comment by Nicolas Flagey on JIRA:

Mosaic obtained with pipeline without our own background matching: https://stsci.box.com/s/h486zcatdcrrq4i5bh3t5eew0lxqgfbq

Mosaic obtained with pipeline after our own background matching: https://stsci.box.com/s/3sillvgnnblobxqik7hrmals06g7n8ez

These two mosaics only use the NIRCam B3 and B4 cals but if needed, I can also send you the links for the entire field.

[stscijgbot-jp]

Comment by David Law on JIRA:

Looks like I can't access any of those box links?

[stscijgbot-jp]

Comment by Nicolas Flagey on JIRA:

Can you access this folder: https://stsci.box.com/s/zj8te9xb1447gjuqovgabqmpj01gagt7 ?

And then check for the mosaics with a filename that contain B3B4?

[stscijgbot-jp]

Comment by Nicolas Flagey on JIRA:

I send direct links to the mosaics to Ned Molter, David Law and Mihai Cara via email

[stscijgbot-jp]

Comment by David Law on JIRA:

I can't access the folder, but I can now see the individual mosaics linked above.

[stscijgbot-jp]

Comment by David Law on JIRA:

Looking at the data I assume it's getting thrown off by the bright diffraction spike from the 4th mag star nearby, but couldn't say why offhand.  Will be useful to hear Mihai Cara 's opinion on this case.

[stscijgbot-jp]

Comment by Nicolas Flagey on JIRA:

David, that might be true, but why is it that the code we run off the pipeline works then? My understanding is that the pipeline is doing the exact same thing but fails. The devil might be in the details here, about how the overlap regions between CALs is computed and what kind of filtering, if any, is done ...

[stscijgbot-jp]

Comment by Mihai Cara on JIRA:

It would be very helpful if I could somehow get access to the data and possibly a notebook/script indicating how stage 3 was run (skymatch, tweakreg, resample) so that I could reproduce the original ({}"Mosaic obtained with pipeline without our own background matching"{}) resampled image. 

[stscijgbot-jp]

Comment by Ned Molter on JIRA:

Hi Mihai, these should all be provided already.  See the "data location" field at the top of this Jira ticket for the data in a Box folder, and see the attachment to this Jira ticket for a notebook showing how Nicolas ran the custom match algorithm

[stscijgbot-jp]

Comment by Mihai Cara on JIRA:

Nicolas Flagey Please see attached screenshot.

[stscijgbot-jp]

Comment by Mihai Cara on JIRA:

Maybe you could put these files on central store with appropriate permissions for me to get access to them?