Skip to content
GitLab
Projects Groups Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
  • Sign in
  • athena athena
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
    • Locked Files
  • Jira
    • Jira
  • Merge requests 138
    • Merge requests 138
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Releases
  • Packages and registries
    • Packages and registries
    • Container Registry
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Code review
    • Repository
  • Activity
  • Graph
  • Jobs
  • Commits
Collapse sidebar
  • atlasatlas
  • athenaathena
  • Merge requests
  • !26648

unflipped DL1(r) evaluated with (some) flipped input variables

  • Review changes

  • Download
  • Email patches
  • Plain diff
Merged Philipp Windischhofer requested to merge phwindis/athena:21.2.73.0_DL2_JetFitter_flipped into 21.2 Sep 17, 2019
  • Overview 14
  • Commits 4
  • Pipelines 0
  • Changes 2

Hi,

This MR fixes a bug that shows up when both flipped and unflipped versions of DL1(r) are scheduled. Currently, the input variables (minimum|maximum|average)TrackRelativeEta are not stored separately in the BTagging container for the flipped and unflipped versions of the tagger (see here). Since the BTagJetAugmenter for the flipped tagger is scheduled after the one for the unflipped taggers (see here and here), the input variables for the latter are overwritten and DL1(r) is always run on (minimum|maximum|average)TrackRelativeEta belonging to the flipped version.

The fix implemented here follows suggestions by @dguest and seems to work fine at first glance. More testing is probably desirable.

Note that this problem affects derivations that schedule the flipped taggers, i.e. all FTAG-derivations. Physics derivations without the flipped taggers should not be compromised. This means that the FTAG derivations at the moment contain a strange variant of DL1(r), which is different from the one in the physics derivations and also different from the tagger that was actually trained. I guess this is not really want we want for the next round of calibrations.

Any feedback is appreciated!

Cheers, Philipp

CC: @guirriec @cschiavi @cpollard @lidiaz @sanmay

Edited Sep 18, 2019 by Philipp Windischhofer
Assignee
Assign to
Reviewers
Request review from
Time tracking
Source branch: 21.2.73.0_DL2_JetFitter_flipped