DM-39243: Add an initial cell-based coadd task #33
Conversation
arunkannawadi
force-pushed
the
tickets/DM-39243
branch
from
e7fe45b to
b64c165
Compare
| for cellInfo in skyInfo.patchInfo: | ||
| stacker = AccumulatorMeanStack( | ||
| shape=(cellInfo.outer_bbox.getDimensions().x, cellInfo.outer_bbox.getDimensions().y), | ||
| bit_mask_value=statsCtrl.getAndMask(), |
There was a problem hiding this comment.
If this is all we're using from StatisticsControl, I'd be inclined to drop it and reimplement this method in Python.
There was a problem hiding this comment.
We still pass the StatisticsControl object to _compute_weight above that uses a clipped mean of the variance plane to compute the weight.
There was a problem hiding this comment.
I've replaced this specific use of statsCtrl with the line that set its AndMask in the first place. However, statsCtrl in the code is to stay for now, until we can robustly estimate weight in a later ticket.
| ) | ||
| badMaskPlanes = ListField[str]( | ||
| doc="Mask planes that, if set, the corresponding pixel would not be included in the coadd.", | ||
| default=("BAD", "EDGE", "NO_DATA", "SAT"), |
There was a problem hiding this comment.
What does this actually mean for cell-based coadds, where we need to use a consistent set of epochs for all pixels in the coadd? I thought we'd have to interpolate mask planes like these on the warps (here, after reading them, would be fine), and then just add the interpolated values as if they were real (after thresholding on the number of such pixels, etc.).
The implementation seems to happen inside AccumulatorMeanStack (and I'm not familiar with its innards), so is this something the DESC coadds are also doing?
If we do end up dropping this option - or using it to interpolate warps after reading them in - I'm not sure we'd be getting much out of AccumulatorMeanStack, especially after we start populating SingleCellCoadd.inputs, since that will given us another accumulated-as-we-go data structure with weight information we can divide by later. (We certainly don't have to take this step now.)
There was a problem hiding this comment.
These are the mask planes that should count towards the calculation the fraction of masked pixels. I updated the doc to reflect this. This, and max_maskfrac are currently just placeholders until we do the interpolation, which we don't right now.
The DESC coadds use AccumulatorMeanStack as well, but the interpolation does not happen within it. The DESC code interpolates the bad pixels on the calexp before warping and pass them onto the AccumulatorMeanStack.
arunkannawadi
force-pushed
the
tickets/DM-39243
branch
from
d84cf27 to
37ef432
Compare
arunkannawadi
force-pushed
the
tickets/DM-39243
branch
from
37ef432 to
4010cf0
Compare
arunkannawadi
force-pushed
the
tickets/DM-39243
branch
3 times, most recently
from
da4b5d7 to
9c29655
Compare
arunkannawadi
force-pushed
the
tickets/DM-39243
branch
from
9c29655 to
7dfee69
Compare
This does a bunch of refactoring to the existing
AssembleCoaddandCompareWarpAssembleCoaddtasks to switch allow for old type of coadds and the new cell-based coadds. The task constructs the coadds using warps and does not use any DESC code for now. This places the infrastructure in place to make further developments.Edit: The old refactoring code described above has been punted to DM-41632. This is clean(er) stand-alone pipeline task without relying on the legacy CoaddBase code base.