Fix ND-LAr reco related hacks

chenel commented 1 year ago

In order to finish testing out ND-LAr ML-Reco branch filling w/ truth matching, there were a number of short-term workarounds necessary for compatibility with MiniRun-3 generation files. These will need to be addressed when upstream fixes land:

[x] vertexIDs are rewritten in edep-sim to be consecutive and so are not able to be uniquely matched to GENIE events in the .ghep files without using an intermediate file. (This is particularly problematic for rock neutrino events.) This will be fixed in files after MiniRun-4. A workaround that searches for matching particle content (at present: primary lepton energy) is used instead, but this is fairly inefficient.
[x] Currently the truth information reformatted Supera used by ML-reco tracks only particles, so information unique to neutrino interactions (for example: their vertexID) is lost. A workaround using true primary lepton energy is used to match events, but some information is still irretrievably lost. Work is in progress to add the necessary Supera pass-through class and to fill it using flow2supera.

There's also a separate problem that's not a hack per se but which also needs to be sorted out:

[x] The is_primary field for truth_particles is not being deserialized correctly from ND-LAr reco files. It appears to be stored correctly, and when I view the files with h5py I see both true and false values, but via the C++ interface I'm always seeing false.
[x] Double the number of entries filled in common.ixn.dlp
[x] Double counting of common.ixn.ndlp
[x] Fill sr.nd.lar.ndlp

chenel commented 11 months ago

Seems the is_primary thing has fixed itself?

chenel commented 11 months ago

We don't actually need all the true neutrino info via pass-through if we can correctly propagate the vertexID, so long as we're using GENIE matching. (All the true neutrino info we'd ever want is in GENIE.) So I'm going to back-burner that component for now and focus on the GENIE matching.

chenel commented 9 months ago

We've finally got the vertexID stuff in place with #28, so we can close this one out.

DUNE / ND_CAFMaker

Fix ND-LAr reco related hacks #24