Large Radius Tracking optimization: Include nn cut in ambiguity solver
This MR includes changes to the InDetAmbiScoringTool and associated files to allow for the rejection of fake tracks using a neural network discriminant. The default configuration of this cut is set to below 0 (i.e. off) and therefore has no effect on the current large-radius tracks unless explicitly set. This tool is configured to have no effect on standard tracking.
Inclusion of this network was discussed in Tracking CP and approved for testing as we continue to develop features of the R3 LRT optimization effort.
Please see this LLP forum presentation for details on the network and performance. See this Tracking CP presentation for an overview of the LLP optimization effort to date.
The NN configuration is located in /cvmfs/atlas.cern.ch/repo/sw/database/GroupData/dev/TrackingCP/LRTAmbiNetwork/nn-config.json
and retrieved with the PathResolver
tool. For inclusion in a release, this configuration file will be moved to a more permanent area (not dev/).