Investigating ITS-TPC matching (in)efficiency
- Selected were primaries with NCL > 60. Problem: Lower efficiency for NN regression algortihm, independent of cluster classification threshold.
Investigation of NN and CF regression performance
- CF regression finds more TPC tracks (0.15%) while NN regression produces more clusters (0.52%)


- CF[60:]-NN[60:] = -609 -> NN finds more tracks with NCL > 60
- CF[:60]-NN[:60] = 8955 -> CF finds a lot more short tracks that are not used in physics
- NN regression potentially reduces NCL for tracks with high NCL. For large clusters, a more accurate CoG will be "further away". Such clusters are actually bad -> get excluded in dE/dx calculation by truncated mean

Back to ITS-TPC matching:
- Number of found matches: 1.3% less matches for NN (global, no selection)
-