psheehan
commented
1 year ago Ok, the problem was that some of the changes for the fix to make sure that the model was properly applied to all relevant data had some places where the same lines were changed as this update. I've merged those properly and this should no longer have any conflicts.
jjtobin
Initially auto_selfcal was using the model from the _pre image to more quickly generate the _post image after calibrations were applied. This was computationally more efficient, but could occasionally lead to poor subtractions and large residuals in the _post image. This merge has the code now running a full tclean for the _post image. To get back some of the computational efficiency, it uses the _post image as a starting point for the _pre image for the next solution interval. (This was initially an option that could be turned on, but has now been made standard - is this ok?).
One quick question to check - it should always be the case that the threshold for a subsequent solint is <= the threshold for the previous solint, right? Want to avoid cases where somehow the threshold goes up and so the _pre image is over cleaned compared with the _post image, if that could happen.
Github is also telling me this can't be automatically merged, not entirely sure why but will have to sort that out. In the meantime, the full_tclean_post branch in my fork can be used for testing.
Mentioning @r-xue in case this becomes relevant for the pipelin version.