Skip to content

Improve truth vertex labelling

Samuel Van Stroud requested to merge svanstro/training-dataset-dumper:r22-improve-truth-vertex-indexing into r22

!150 (merged) added the truth production vertex index for each track. Upon closer inspection we can improve this labelling:

  • Find the PV, call anything within a millimetre of the PV a PV vertex. PV gets an index of 0.
  • For other vertices, check if we have a vertex with 1mm, if so, just reuse that.

Example output for a b-jet:

  • Before change: [450 781 780 781 779 457 449 454 451 458 449 449 452 449 766 -1...
  • After change: [ 2 1 1 1 1 0 0 0 0 0 0 0 0 0 1 -1...
  • Track origins (3 = B, 4=B->C, 2=Primary): [ 3 4 4 4 4 2 2 2 2 2 2 2 2 2 4 -1...

Before we had many close-by vertices with different indices. Now we merge these, hopefully giving a clearer signal to train a vertexing alg.

Is this the best way to do this? Is the 1mm cutoff reasonable? Maybe there is a more informed approach from SV.

@jshlomi

Edited by Samuel Van Stroud

Merge request reports