Models glitch loader off of ml4gwdataloading tools.
Implementation would be cleaner if instead of a directory of files for each ifo, we had individual files containing datasets of glitches for each ifo. I think I'll leave this to a future PR since this is working.
Throughput is more or less equivalent to before (~1400 samples per second on dgx boxes).
setting glitch_frac = 0.0 should turn of glitch sampling and reproduce previous behavior (will add test for this)
@wbenoit26 Let me know if you have any feedback. Going to launch a full pipeline run with an ad-hoc glitch_prob choice to ensure things don't break and results look reasonable.
Reworked version of #367.
Models glitch loader off of
ml4gw
dataloading tools. Implementation would be cleaner if instead of a directory of files for each ifo, we had individual files containing datasets of glitches for each ifo. I think I'll leave this to a future PR since this is working.Throughput is more or less equivalent to before (~1400 samples per second on dgx boxes).
setting
glitch_frac = 0.0
should turn of glitch sampling and reproduce previous behavior (will add test for this)