chime-experiment/ch_pipeline

flagging warning over MPI ufunc

Closed this issue · 3 comments

timestream.rebin_weight[:] *= weight_mask[:, 0]

In the @ljgray lg/rebin-weight-masks branch, there is some scary rebin_weight action that dumps hundreds of warnings when I generate quarterstacks:

  9979.2s [MPI 122/128] - WARNING  py.warnings: /project/6003614/zackli/pipeline2/chime_env/code/ch_pipeline/ch_pipeline/analysis/flagging.py:2122: UserWarning: A ufunc is combining an MPIArray with a numpy array that matches exactly the local part of the distributed axis. This is very fragile and may fail on other ranks. Numpy array shape: (8, 4096); dist axis: 0.
  timestream.rebin_weight[:] *= weight_mask[:, 0]

I converted the sstack of the rev_08 fullstacks with the VisGrid task and generated some ringmaps that looked super weird — I'll do some digging but this might be part of it?

Ah, yeah the way distributed arrays are implemented will throw a warning, but the behaviour is still ok. I'll fix the warning in the morning, but it won't affect weirdness with ringmaps

I'd be curious to see the ringmap weirdness

I'm going to close this issue because we've gotten rid of the `rebin-weight-masks1 branch