Skip to content

Update of large-R jet DNN calibration tool - handling specific cases

Guillaume Lucas Albouy requested to merge galbouy/athena:largeRDNNCalib_exc into main

This MR introduces specific treatments in large-R jet DNN calibration for particular cases:

  • jets with null or negative masses, jets with only one constituent, both causing problems for DNN inference --> jets are left at the default constituent scale (l. 319-322)
  • jets with masses below 40 GeV for which the mass calibration is not optimal --> only the energy calibration is applied for these jets (l. 387-397)

These specific cases and their treatments where discussed during the last JSV meeting.

MR changes also the default values of Tau21_wta and Tau32_wta to -0.1 to match what is done during DNN training. (@cdelitzs, @delsart)

Secondary: adds some new comments and re-indentation, the last one messes up a bit the git diff.

Tagging jet conveners @fballi, @camacho, @mswiatlo, @bhodkins, @pereirap, @haweber.

Merge request reports