Multi-track fitting porting

[brettviren]

The "multi-track fitting" portion of WCP is to be ported to WCT.

We must map some new data types to WCT

• ProtoVertex
• ProtoSegment
• WCShower

In-source notes here: https://github.com/WireCell/wire-cell-toolkit/tree/apply-pointcloud/patrec/docs/tjft

Some "proto-proto" versions of these types are already needed for #435. While multi-track fitting is later in the processing chain, we will try to provide these data types first so they may be used for that earlier stage.

[brettviren]

Here is a UML diagram showing the neighborhood of these three WCP types.

• Point is essentially the same type with this name in WCT
• TVector3 can be replaced with WCT's Vector (which is just Point)
• The two Map_Proto_X_Ys are map<X, set<Y>>.

Some basic data pattern observations

• All have a "bag of data" part
• The Map_'s represent a bipartite graph connecting vertex and segment graph nodes/vertices.
• Shower and segment hold references to point cloud objects

Questions:

The point clouds

• What are the set of points in the segment's PCs?

There are two "associated" PCs (nominal and steiner) and one "fit" PC.
The points are provided by the creator / caller of the segment, eg WCP's NeutrinoID.

The points in the "associated" PC are a subset of those in the, err, associated PCs (nominal or steiner) of the cluster.

The "fit" PC and the fit_pt_vec vector go together and are both set at construction but may be both reset later (set_point_vec() or set_fit_vec()).

• Are they relevant only during construction of the segment or shower instances?

Maybe both(?) but seem like they need to be part of a WCT version. These PCs all need k-d trees.

Graph representation?

• Is an explicit graph representation of the relationship between ProtoVertex's and ProtoSegment's useful instead of replicating the two Map_X_Y?

• If yes, should that graph be bipartite with vertex+segement node types?

• Or, should the graph be point-level with each point being distinguished as being a "ProtoVertex" point or a "point in a ProtoSegemetn".

I note that the break_segment_at_point() operation is a more natural fit to a point-level graph.

What is equivalent to "mcell"?

• WCPoint has pointer to mcell (WCT's IBlob). Is that needed? Is it enough to hold a reference to a PC Tree blob node Blob facade?