How do we support the many heuristic clustering functions?

[brettviren]

Heuristic clustering functions.

They will:

• mutate a PC
• generally fast operation
  • do not want overhead of making full flow-graph-nodes sharing IData
• require configuration
• many get combined in a pipeline pattern
  • some short but unrollable loops
  • pipeline depends on detector, needs to be configurable

Nominal solution idea:

• Follow "tool interface" pattern
• Define an interface, eg IPointCloudMutator::mutate(PCDataType& pc)
• Define a number of concrete implementations
• They may also be IConfigurable
• Implement a special IPointCloudMutator which is configured with a pipeline of others
• Implement a DFP graph node that builds initial PC, configured with one IPointCloudMutator to process it.
• Port the dozens of WCP cluster heuristic functions, each into a IPointCloudMutator

[brettviren]

Some brief heuristic descriptions.

• project 3D position (2D transverse position) to 2D wire view (1D pitch location)
• find channel info associated with a point
• judge if point is "good" or associated with dead channel(s)
• estimate charge from 2D projection and 3D views
• identify dead channels "near" to a point
• cluster "live" and "dead", "jumping gaps"
• "cluster extend with strategy" (prolonged, parallel, regular, close)
• "clustering separate" judge if a cluster should be broken up
• "cluster deghosting"
• "clustering examine boundary" fix when clusters are larger than physical
• "cluster connect" many small disconnected clusters in a line
• "clustering protect over clustering"
• "clustering neutrino"