Improve eval plots by using simulation information in target particle clustering #370
Description
We have evaluation plots such as these:
- https://huggingface.co/jpata/particleflow/blob/main/clic/clusters/v2.0.0/plots_checkpoint-29-1.901667/clic_edm_ttbar_pf/jet_res.png
- https://huggingface.co/jpata/particleflow/blob/main/clic/clusters/v2.0.0/plots_checkpoint-29-1.901667/clic_edm_qq_pf/jet_res.png
- https://huggingface.co/jpata/particleflow/blob/main/clic/clusters/v2.0.0/plots_checkpoint-29-1.901667/clic_edm_ww_fullhad_pf/jet_res.png
In these plots, the targets and truths have an inherent smearing that is due to jet clustering of the target particles.
Since we know which target (interacting status=1 and status=0 particle) came from which truth particle (all status=1), we could get rid of the smearing from clustering by using the target-true matchings in the jet. That means that we cluster the target particles based on the originating truth particle index, rather than based on the pt,eta,phi of the target particles.