DM-54482: Integrate PSF-shapelet subtask into calibrateImage as a donut-detector#471
Conversation
We have determined that it is necessary to use an order of 6 for the PSF star shapelet expansion for current LSSTCam data, so this should be the default.
| shapelet_summary.nShapeletStar = len(shapelet_used_cat) | ||
| centroid_diff_shapelet_vs_slot = np.sqrt( | ||
| (shapelet_used_cat["slot_Centroid_x"] - shapelet_used_cat[self.shapelet_base_name + "_x"])**2.0 | ||
| + (shapelet_used_cat["slot_Centroid_y"] - shapelet_used_cat[self.shapelet_base_name + "_y"])**2.0 |
There was a problem hiding this comment.
I think this is the only point at which ComputeRoughPsfShapeletsTask requires SingleFrameMeasurementTask to have been run already. What happens right now when we have a failure during measurement (especially an all-centroids-failed error)? I don't really have any great ideas for how to deal with this, but while I like including the centroid offset in the score most of the time, I don't like the idea that this stops us from computing the score when we can't centroid now.
There was a problem hiding this comment.
I’ve yet to hit this case, but it’s a good point! The only idea I have at the moment is to define two shapelet scores; one that only includes the non-atmospheric power and one also including the centroid offset whenever possible (and the latter could be set to nan if the centroiding in failed).
There was a problem hiding this comment.
Yeah, that's what I was coming around to as well. Sorry to give you another field to add, but I think that's what we ought to do.
There was a problem hiding this comment.
No probs. Any thoughts on naming? I was thinking of shapelets_score_non_atm & shapelets_score, with the latter being the “authoritative” one when available (I’m not sure if it’s more or less confusing to try to add more info to the name of the latter…)
There was a problem hiding this comment.
How about shapelets_only_iq_score and shapelets_iq_score?
There was a problem hiding this comment.
Ok, I've implemented a second method to compute the shapelets_iq_score after the measurement run of the PSF catalog. In doing so, it occurred to me why I never his this failure mode...I don't actually think it's possible. The reason being that the SDSS centroid algorithm has the config:
# If >0; maximum distance in pixels between the footprint peak and centroid allowed before resetToPeak flag is set.
config.star_measurement.undeblended['base_SdssCentroid'].maxDistToPeak=1.0
which, for calibrateImage is set to 1 pixel. So, if the centroider would like to wonder more than one pixel from the original peak, it reverts back to that peak. So, while this is flagged and hence formally a centroiding failure, the slot_Centroid_x & slot_Centroid_y columns always have an entry. I have not been checking for centroid flags in any of this since they are expected to be set for very non-Gaussian shapes (i.e. funky PSFs).
I do think it's a good idea to include both "scores", but I'm curious what your thoughts are on the above...
There was a problem hiding this comment.
Good point. I think there might be some very narrow room for NaN centroid columns - i.e. failures before the centroider runs - but I'm not surprised we don't see any of those in practice right now. But I'm also in favor of keeping both scores to hedge against that.
laurenam
force-pushed
the
tickets/DM-54482
branch
2 times, most recently
from
d74ad86 to
3c519a5
Compare
laurenam
force-pushed
the
tickets/DM-54482
branch
from
3c519a5 to
a86c70e
Compare
2 participants
No description provided.