Problem in Data reading stage & Graph Building stages with new samples (v5: Only hits which are not tagged as overlaps in phi are selected)
In v5 events only hits which are not tagged as overlaps in phi are selected. So the hit ids won't be consecutive, we will have some "holes" in the hit ids.
When I evaluate the MM-based graphs construction efficiency on 10 v5 seen events, here is what I got:
.
If we follow a particle which have missing edges in the graph, let say particle 3560010585 in event00000001 here is what we got:
Expected tracks : 108397 -> 116399 -> 124223 -> 131958 -> 96701 -> 92568 -> 310777 -> 314276 -> 283330 -> 288226
Track edges constructed by acorn before remapping:
108397 -> 116399 -> 124223 -> 131958 -> 96701 -> 92568 -> 310777 -> 314276 -> 283330 -> 288226
Track edges after remapping:
108397 -> 116399 -> 124223 -> 131958 -> 96701 -> 92568 -> 309516
Edges in the graph after graph building (according to truth map):
108397 -> 116399 -> 124223 -> 131958 -> 96701 -> 92568
There is a conceptual problem because after remapping, in the case where the hit_id are not consecutive we will have "artificial" numbers in the track edges which are not hit_ids any more but we still construct the edges in the graph from pairs of hit_ids so when do the intersection between track_edges and edge_index it will mess up :
Same for evaluation, obviously...