Skip to content

More fixes for full retagging

Dan Guest requested to merge dguest/training-dataset-dumper:retag-fix into main

This is hopefully enough modification to the full retagging to get @vhomsak's studies going.

The main changes:

  • Removed a few arguments to the script, specifically those that need to be in sync with the config file anyway. Now we grab them from the config file instead.
  • Always run the flip taggers, it's no longer an option.
  • Always run jet calibration before tagging, and optionally some systematic. Note that this isn't how things are done in production, but there are no systematics for pre-calibration jets so we have to do things slightly differently here.
  • I added the flip taggers to the output.

As is always the case with "retagging"

  • Everything is very fragile and confusing
  • It's slow to debug
  • I don't know for sure if it's right

So while this is probably good to go I would not trust that it's correct without some validation.

Note: in an earlier iteration of this MR I wasn't able to get the jet systematics to work. The short explanation is that I was confused by the interaction between the address remapping service and shallow copies. Looking back in history, this was working correctly a few years ago, but any jet systematics in retagging probably weren't doing anything since !499 (merged) went in in May 2023.

Edited by Dan Guest

Merge request reports