Non natural weighted sums & tests on LP212-62

cyriltasse / DynSpecMS

0 stars 3 forks source link

Non natural weighted sums & tests on LP212-62 #17

Open cyriltasse opened 4 years ago

cyriltasse commented 4 years ago

So far I've modified DDF to save the imaging weights, and also updated DynSpecMS to take them into account. Will post the results as they come.

cyriltasse commented 4 years ago

Ok so here are the results on L230461

Natural:

Briggs -0.5

Yet I'm surprised, doing a residual with the final DicoModel, properly taking the DD solution files, and from the MSs Tim has given me L230461_SBgr*-10_uv.MS.pre-cal.ms I get something quite crapy (DATA column only present) with an rms noise of 4 mJy !!!

twshimwell commented 4 years ago

The bottom one is an image of the who field is it? What is the command you are using to make it?

cyriltasse commented 4 years ago

Yeah it's a residual image made with

DDF.py image_full_ampphase_di_m.NS_SUB.parset --Data-MS mslist-L230461.txt --Cache-Reset 1 --Output-Name TestWeights --Weight-OutColName DDF_WEIGHTS --Parallel-NCPU 56 --Output-Mode Dirty --Predict-ColName DD_PREDICT

cyriltasse commented 4 years ago

I realised the column was WEIGHT_SPECTRUM in the parset, any reason for that? I'll try using IMAGING_WEIGHT, see if that helps...

twshimwell commented 4 years ago

Ah right that might be the cause. Not sure why its different, perhaps when we archive we change the weight column name but i cant remember...

cyriltasse commented 4 years ago

ok I'm having a go... could be a big help to get something better...

cyriltasse commented 4 years ago

Indeed, the residual image gets way better..

we couldn't take any weights into account before, let's see how that difference plays with the dynamic spectra...

twshimwell commented 4 years ago

Oh great

cyriltasse commented 4 years ago

Ok so here is the dynamic spectra with and without the kMS weights being applied:

and without (original)

The difference is quite noticable...

jrcallingham commented 4 years ago

Interesting! Is it my eyes or is the centre of clux of the bright blob on the right shifted relative to the two spectra? This makes some sense why matching between polarisations was noisy

cyriltasse commented 4 years ago

It's not exactly at the same place, I've just selected the region with the mouse, so no the two spectra are offset... the ripples do seem less strong in the weighted one...

twshimwell commented 4 years ago

the weights here are a bit odd right. We are just looking at the relative weights between the different baselines? Because if e.g. some timeslot at some freq is downweighted equally on all baselines because its dodgy (or has some flaring star in this case) then it wont change the dynspec