This MR schedules the new flavour tagging algorithm trainings for PFlow jets and VR track jets.
The training of a graph neural network using the GN2 architecture with a 190M jet sample for PFlow jets is deployed. Also, trainings for VR track jets have been deployed, both for the DL1dv01 algorithm (including a dedicated DIPS training which is required for DL1d) and for the GN2 architecture but trained with a training sample of smaller size than the PFlow version.
The new trainings are:
- PFlow GN2v00:
BTagging/20230306/gn2v00/antikt4empflow/network.onnx - VR track jets DL1dv01:
BTagging/20230307/DL1dv01/antiktvr30rmax4rmin02track/network.json - VR track jet DIPS (loose track selection):
BTagging/20230208/dipsLoose/antiktvr30rmax4rmin02track/network.json - VR track jets GN2v00:
BTagging/20230307/gn2v00/antiktvr30rmax4rmin02track/network.onnx
In addition, the provisional trainings which were based on PFlow samples and were scheduled for VR track jets to fill the gap, have been removed now that dedicated algorithms for VR track jets are available.