Read pyuvdata-formatted beamfits beam models #301
Conversation
|
An example beam fits file is available here: https://github.com/rlbyrne/rlb_LWA/blob/1db84dcbec1071b2039886c48130047bc1009b1d/LWAbeam_2015.fits |
rlbyrne
force-pushed
the
uvbeam_support
branch
3 times, most recently
from
5c307c0 to
ca206f3
Compare
|
I've implemented a memory reduction for the imported beamfits that is present for normal beam generation (changes two lines of code). Memory usage is below 80G for a typical MWA run. Typical MWA runs of this type are below 60G. However, in the memory usage profile of the job, there is a spike during beam generation. If I can find that, it can be just as efficient as a normal run. The resulting image is wrong, however. I'm using a beamfits of the MWA produced by pyuvdata. Looks like the beam (right) is 1's within the horizon (data on left). This could be because of the beamfits I've produced, or could be a problem of the code. More investigation is required (advice welcome). |
|
@nicholebarry Thank you for the memory speedup! I'm afraid I don't understand the images you posted. How are you finding that the images are wrong? I would expect that the pyuvdata beam wouldn't be exactly equal to the MWA beam contained within FHD. |
|
Unfortunately the "Reduce mem" commit seems to have introduced a bug. Here is an example of an image made without that commit: |
|
Fixed the problem with the mem reduction. There is now some push towards combining some of the beamfits code as pyuvdata_beam_import is quite small. @rlbyrne can you check that the change does not affect LWA data? I need to rerun, because the beam is ever so slightly off, but I didn't include the recent commits. So I will set that up now and have it finished soon. |
|
@nicholebarry Thank you. I will run an LWA test. Please let me know what you find with your test. |
|
I just tested the latest version of the branch on LWA data and the results look good. @nicholebarry are there any more tests we need to run? |
This PR allows the user to specify the beam model by supplying the path to a *.beamfits file.