Muon learning

• Deployment of the technical infrastructure to apply machine learning for phase-II muon reconstruction on MuonSpace points. The package MuonInferenceInterfaces is the foundation of the entire ML infrastructure. The idea is that all ML algorithms run on the same graph structure which is handled by the InferenceAlg. The particular ML networks are hosted each by an IGraphInferenceTool. The GraphRawData represents the cache to align the input data for the ML-network and also to build the final graph interfaced to the ONNX library. The GraphInferenceToolBase implements the basic functionallity to load the ML from disk and also to build the graph from the MuonSpacePointContainer. To allow for flexibility in the development chain, the names of the used features are saved to the network's metadata and then later mapped to the NodeFeatureFactory. The glue between the factory and the feature names is the NodeFeature class.

The MuonSPID package serves for validation purposes of the ML response and won't be used later in the reconstruction jobs. Follow-up MRs are already to foreen to streamline the connection between the graph nodes and also to establish a full muon reconstruction chain.

Tagging: @jojungge, @atlasbot, @minions, @markowen, @emoyse