residual structure being left in image

[o-smirnov]

It seems some recent changes are making PyMORESANE quit the deconvolution at a shallower level. Something for you to explore @JSKenyon when you get your teeth sunk into the pipeline.

For example, see recent pipeline results here: /net/aretha/home/oms/VLA-CygA/reduction-18Aug15-HI-224chan-chan16-lsmthr0.25-spi-moresane3-rob-2-multisig/*/*s7*.res*

I'm looking at the stage 7 image produced like this:

wsclean -niter 3 -scale 5.55555555556e-05 -superweight 4 -weight briggs -2.00 -moresane-arg '-sbr 1024   --negcomp' -channelrange 16 240 -moresane-sl 12,6,3 -datacolumn CORRECTED_DATA -name reduction-18Aug15-HI-224chan-chan16-lsmthr0.25-spi-moresane3-rob-2-multisig/plots-CYG-S-HI-128-224/CYG-S-HI-s7.moresane -field 2 -pol IQUV -gain 0.05 -moresane-ext /home/oms/PyMORESANE/pymoresane/bin/runsane -threshold 0 -nosmallinversion -size 2048 2048 -mgain 0.85 CYG-S-HI.MS '

(Note the new -moresane-sl 12,6,3 option: meaning, use --sigmalevel 12 on the first major loop iteration, 6 on the second, 3 on the third. Except I just found my wsclean patch was broken, bah, it was using -sl 3 throughout... should be fixed now).

By contrast, the stage 7 images from a month ago look like this: /net/aretha/home/oms/VLA-CygA/ref/reduction-23Jul15-HI-224chan-chan16-lsmthr0.25-spi-moresane3-rob-2/*/*s7*.res*

The command was:

wsclean -niter 2 -scale 5.55555555556e-05 -superweight 4 -weight briggs -2.00 -moresane-arg '-sbr 1024  --sigmalevel 3 --negcomp' -channelrange 16 240 -datacolumn CORRECTED_DATA -name reduction-23Jul15-HI-224chan-chan16-lsmthr0.25-spi-moresane3-rob-2/plots-CYG-S-HI-128-224/CYG-S-HI-s7.moresane -field 2 -pol IQUV -gain 0.05 -moresane-ext /home/oms/PyMORESANE/pymoresane/bin/runsane -threshold 0 -nosmallinversion -size 2048 2048 -mgain 0.85 CYG-S-HI.MS 

As you can see, this just has an explicit --sigmalevel 3. So, both images have effectively been deconvolved down to 3 sigma -- but the one from last month has much lower residual.

[o-smirnov]

For reference, here are the images. Check the recent "s7" residuals:

...versus last month's "s7" residuals: