lwimager fixed=1 niter=0 operation=csclean no longer fills MODEL_DATA

[o-smirnov]

In version 1.2, we had the wonderful feature of being able to use "fixed=1 niter=0 operation=csclean" to populate the MODEL_DATA column with visibilities computed from an pre-existing model image.

(pyxis imager.predict_vis takes advantage of this, for example.)

As of the current version this no longer fills the MODEL_DATA column (although the status messages indicate everything is going on as usual).

Furthermore, telling it to continue clean from a fixed model (niter=1000 for example) causes the whole thing to bomb out ingloriously:

(/build/buildd/lwimager-1.3.0/synthesis/MeasurementComponents/CSCleanImageSkyModel.cc : 467) Failed AlwaysAssert imagePtr

[o-smirnov]

Possible solution to first problem: create Imager object with useModel=True at this line:

https://github.com/ska-sa/lwimager/blob/4df398bd0879cae0153931bfec531ef5355f0530/synthesis/apps/lwimager.cc#L487

[o-smirnov]

Fixed this -- note that I had to take out a conditional in CSCleanImageSkyModel as well (somebody seems to have explicitly killed this functionality, naughty naughty!)

[IanHeywood]

Perhaps related to CASA's clean task no longer storing the inversion of the clean component model in the MODEL_DATA column by default...

[o-smirnov]

Yeah it is related I guess -- they have a usescratch keyword for that in the task, and the default is False. I've added a fillmodel option to lwimager as of this fix (will show up in 1.3.1 release), use fillmodel=True to make sure the MODEL_DATA column is filled.

[IanHeywood]

I guess all the hollering about the MS format being a data hog finally found some traction. CASA workaround is the ft task with usescratch=True, for the record.

[o-smirnov]

Ahh that's good -- I plan to wrap CASA in pyxis next (to help preserve my sanity), this sort of stuff is exactly what I need to know!

[IanHeywood]

It already works pretty well from within Pyxis via tricks like this:

def imageMS(msName,imageName,dirty,fixedbeam):
    cc = 'clean('
    cc += 'vis="'+msName+'",'
    cc += 'imagename="'+imageName+'",'
    cc += 'mode="mfs",'
    cc += 'gridmode="widefield",'
    cc += 'wprojplanes=64,'
    if dirty:
        cc+='niter=0,'
    else:
        cc += 'niter=2500,'
    cc += 'imsize=[2240,2240],'
    cc += 'cell=["12.0arcsec"],'
    cc += 'multiscale=[],'
    cc += 'weighting="briggs",'
    if fixedbeam:
        cc += 'restoringbeam = ["2.097649468316E-02deg","1.559506416321E-02deg",
"-1.779342346191E+02deg"],' 
    cc += 'robust=0.0)\n'
    cc += 'exportfits(imagename="'+imageName+'.image",fitsimage="'+imageName+'.fits"
,overwrite=True)\n'
    cc += 'exportfits(imagename="'+imageName+'.psf",fitsimage="'+imageName+'.psf.fit
s",overwrite=True)'
    std.runcasapy(cc)
    deleteExtensions = ['.image','.flux','.residual','.model','.psf']
    for ext in deleteExtensions:
        syscall = 'rm -rf '+imageName+ext
        os.system(syscall)

but just being able to call the CASA routines would be neat. Ping me if I can help.

[o-smirnov]

Oops, not quite fixed yet...

[o-smirnov]

@IanHeywood thanks for the ft tip -- it's lightning fast, but I have a nasty suspicion it does not apply w-projection, just from looking at the status messages...

Anyway, this ought to work in 1.3.2 again. In the meantime, let me check if I can get the clean task to do a similarly sensible thing. UPDATE: no I can't, bless you CASA, can't handle IQUV polarized models...