Skip to content

Add NN access for pixel clustering via lwtnn to r22

Katherine Pachal requested to merge kpachal/athena:22.0-add-lwtnn into master

Hello,

This merge request will allow pixel position NN evaluation from jsons stored in a recently added table and using lwtnn, rather than from root files via TTrainedNetwork. This makes it possible to use Mixture Density Networks for the pixel position determination.

At this point, the lwtnn-based evaluation will be turned on by default for position networks. This can be deactivated with a new flag (useNNTTrainedNetworks) to use the older networks; the flag should be used for evaluation of any pre-run 2 sample.

Code has been tested on all relevant scenarios I can think of:

  • Run within lwtnn IOV; flag false -> runs and uses lwtnn/MDN
  • Run within lwtnn IOV; flag true -> runs and uses TTrainedNetwork/old networks
  • Run within lwtnn IOV; flag false; table not in conditions tag -> exits with error
  • Run outside lwtnn IOV; flag false -> exits with error
  • Run outside lwtnn IOV; flag true -> runs and uses TTrainedNetwork/old errors

Tagging @gfacini , @battagl , @npetters , @goblirsc for info.

Thanks for any review comments!

Cheers, Kate

Edited by Katherine Pachal

Merge request reports