handlerDerived open issues

[low-sky]

• [ ] Hybrid masking with velocity priors ("cross scale masking"
• [ ] Maskless data products including Tmax and maybe Vmax?
• [ ] Maskless data products at coarser velocity resolution (12p5kms products)
• [ ] Stacking code check in
• [ ] Continuum subtraction position by position
• [ ] Other issues as assigned

[akleroy]

@low-sky Can you give a quick inventory of whether these are addressed?

Re: "Hybrid masking" - I think this is currently hard-coded, which we definitely need to fix. I will suggest that for next ADR discussion.

Re: "Maskless" - couple thoughts: what Jiayi and I have been doing is using the mask to define a velocity range, so a very, very light mask. It would also be good to get from Annie the algorithm used to make the "most agnostic" maps for the distributions.

I think the first three are implemented, though? Can you update and note open issues.

[akleroy]

Adding and closing other thread:

• [ ] Add cross-scale moments to the "other" moments like vmax. Not clear we would use this in a baseline deployment, but good to have in general. Replicates #81 , which I'm closing.

[akleroy]

• [x] Some mosaics that are built out of bricks have different spectral footprints. The empirical noise generation should account for this so that the noise is not underestimated.

[akleroy]

@low-sky If you want to consolidate issues or organize the thread, go for it. I would expect to be able to help out on this in a few days.

[akleroy]

• [x] Add capacity for lower spectral resolution moments like the 12p5km/s or 5km/s or whatever into the derived product creation.

May already be done. Consolidates #93

[msannih]

Re: "quasi-agnostic masks", for the distributions paper I care create moment-0 maps like this:

(i) use the moment-1 maps at 'low' resolution as a velocity template (I'm using 1kpc I think, but optimal value TBC). (ib) fill in any blanked pixels in the masked moment-1 maps with the value from model velocity field from the dynamics WG

(ii) make a crops dilated mask to identify significant emission in the hires cube and shuffle the mask to the template velocity field

(iii) shuffle cube to template velocity and define a fixed velocity window for integration (I'm using +/-30 km/s, but optimal value TBC)

(iv) logical OR the mask from (ii) and fixed velocity window from (iii)

(v) apply combined mask to shuffled hires cube and integrate

[low-sky]

Thanks! Is the shuffling necessary? Or could we measure a fixed number of channels around the centroid estimate?

[msannih]

shuffling is mechanically how i do it but you're right that specifying a channel range about the channel corresponding to the velocity centroid for each line of sight should be equivalent

[1054]

Adding one issue:

• [ ] when mixing galaxies and lines, if some galaxies do not have some lines, _find_lowest_res will return None and cause exception when passing the resolution to task_hybridize_masks. This is fixed in PR https://github.com/akleroy/phangs_imaging_scripts/pull/150 .

[akleroy]

Some notes:

• [x] Shift broad mask to represent union of all strict / signal masks.

• [ ] In previous version, we added a small velocity-direction dilation to the mask before calculating kinematic quantities. We can reject this but should consider it.

• [ ] There is currently no small region rejection, while there used to be one. This is fine and may be better if we want a clean selection function. We should discuss and resolve this issue.

• [ ] We shifted to the defaults of 2-at-S/N 4. We've used various things over the course of the project, most recently 3-at-3.5 but also 2-at-5. We should discuss whether 2-at-4 is our final default value.

• [ ] The broad mom1 masks had a more complicated recipe previously. At Sharon's request, they imposed a prior and rejected values that deviated from the prior by some threshold. We used a low resolution version of the map as a prior, but circular velocity is also an option. The threshold was pretty mild, about 30 km/s.

• [x] Less crucial, but the ideal interface from my side goes:

handlerDeriver.loop(do_convolution=True, do_noise=True, do_mask=True, do_maps=True)

or whatever. I can do this separation later, but from my side that's what we want to aim for.