search

ratt-ru / meqtrees

A library for implementing radio astronomical Measurement Equations
http://meqtrees.net
10 stars 2 forks source link

Resampler/ModRes problems in a VLA calibration tree #625

Open gijzelaerr opened 10 years ago

gijzelaerr commented 10 years ago
at 2008-03-20 15:49:41 Oleg Smirnov reported:

Resampler/ModRes problems in a VLA calibration tree

gijzelaerr commented 10 years ago

Original comment thread migrated from bugzilla

at 2008-03-20 15:49:41 Oleg Smirnov replied:

I'm currently trying to use resampling with a VLA MS, and have run into all sorts of problems. The MS is here: birch:/data/oms/data/CL0024.MS.

  1. Trying to simulate a few smeared sources using Timba/doc/Courses/Workshop2008/Day5/Supersim.py (use attached finallsm.txt and .tdl.conf), I get empty images. It seems the output of the Resampler nodes contains NANs in the first two timeslots.
  2. Trying to calibrate with smearing included in the model (use Timba/WH/contrib/OMS/Calico/CL0024/calico-vla.py, there's a tdl.conf file in that directory which needs to be copied to .tdl.conf, then run G diagonal solutions) produces two further problems. First, nodes down in the predict tree start falling over with "cells of child results don't match". I have tracked this down to a difference in the time domain of ~1e-6, which makes it fail the comparison at Node.cc line 816. This can be kludged around by increasing the tolerance at Node.cc line 816 (it is currently set to 1e-16), but I'd like to find out where the different time domains arise...
  3. Even with Node.cc kludged to accept the difference in time domains, the solution fails because Resampler does not resample perturbed values properly. If you look at the result of the Resampler's child, it will contain a number of spids and perturbed values, but these are reset to 0 in the output of the Resampler. This unfortunately precludes any resampling in the solve tree...
at 2008-03-20 15:59:58 Oleg Smirnov replied:

Created an attachment (id=46) .tdl.conf for Workshop2008/Day5/Supersim.py

at 2008-03-20 16:00:27 Oleg Smirnov replied:

Created an attachment (id=47) LSM for Workshop2008/Day5/Supersim.py

at 2008-04-28 17:48:12 Sarod Yatawatta replied:

I had a look at this. The reason for all this is the time axis has some zeros in each tile. Do you have any clue how this MS was created? the best solution is to flag all rows with duplicate time values for the same ddid_index.

at 2009-02-23 23:37:15 Oleg Smirnov replied:

Since this is most likely due to a funky MS, I'm moving it off the R1.0 list. I still need to take a look at it later though.

Currently using another VLA MS without problems.