search

CIRA-Pulsars-and-Transients-Group / vcstools

A suite of tools for processing MWA-VCS data
Other
8 stars 8 forks source link

Current development branch is non-determinstic #152

Open robotopia opened 4 years ago

robotopia commented 4 years ago

I've been trying to compare the vdif outputs from different branches that I'm working on to the development branch, and I keep noticing that the results are never identical. I finally tried comparing the outputs of two successive, identical runs of make_beam on the development branch, and the results still differed: the program is non-deterministic. Since there are no random elements being deliberately introduced (that I'm aware of), this counts as a bug (and probably quite an important one, too).

robotopia commented 4 years ago

I need someone to verify what I've found out so far (hence why I assigned Nick). Specifically, a few quick tests seem to show that

  1. only the VDIF outputs are non-deterministic; the PSRFITS files are identical,
  2. it makes no difference whether the analytic beam or the FEE beam is used I need someone else to verify these. At the moment, it's pointing to a bug in the IPFB. Next, I'll be testing out the "determiniticity" of different git versions.