Details
-
Type:
Story
-
Status: To Do
-
Resolution: Unresolved
-
Fix Version/s: None
-
Component/s: afw
-
Labels:None
-
Story Points:2
-
Epic Link:
-
Team:Alert Production
Description
Calib.h has a handful of free functions for converting between physical flux and magnitude (abMagFromFlux, fluxFromABMag) and their errors (abMagErrFromFluxErr and fluxErrFromABMagErr). The former two are used 10 times total outside of Calib and its tests, the latter two are used 8 times (error uses are in PhotoCalTask and fgcmcal). The flux/mag conversions can be trivially replaced by astropy (e.g. (flux*units.Jy).to_value(units.ABmag)), but it's less clear what to do with the error calculations. Unfortunately astropy doesn't support that error propagation directly (though I'm sure they'd be happy to take a PR from us) to do so.
lsst.afw.image doesn't seem like the right place for them to live. Possibly lsst.utils? They'll get more useful names (e.g. nanojanskyErrToABMagErr, abMagErrToNanojanskyErr?) too.
Replacing the flux/mag functions with astropy.units blocks DM-10156.
Attachments
Issue Links
- is blocked by
-
- Done
Activity
After asking around, astropy does have an uncertainty framework which might be able to handle the other case, too: http://docs.astropy.org/en/stable/uncertainty/index.html
Looks like it's only in astropy 3.1 (the stack is on 3.0.3), and is "experimental". It also looks like it wants to do full distributions and estimate uncertainty which is overkill and less stable than when we have analytical unit changes (as in this case):
"Astropy provides a Distribution object to represent statistical distributions in a form that acts as a drop-in replacement for Quantity or a regular numpy.ndarray. Used in this manner, Distribution provides uncertainty propagation at the cost of additional computation. It can also more generally represent sampled distributions for e.g., Monte Carlo calculation techniques."
Replacing the flux/mag functions with astropy.units blocks
DM-10156.
Clearly, it didn't.
While removing the vestigial Calib functions would be nice, it's hard to justify it as something worth spending time on. Closing this as Won't Fix until we have a real requirement for it.
Replacing the flux/mag functions with astropy.units blocks
DM-10156.
Clearly, it didn't.
I'm not sure why these can't live in lsst.afw.image, because that's where PhotoCalib lives, and these routines are clearly related to that. Certainly it's where I would think to look for these sorts of routines.
And (at the risk of repeating myself) I think that if we need/have convenience functions for the errors, we would want convenience functions for the transformations as well, even if they are trivial wrappers around astropy.units.