Skip to content

MVA Models manager

Introducing MVA Models Manager, inspired by Monitoring System ( !1483 (merged)), to simplify MVA models loading:

  • MVA models are initialized inside algorithms, simplifying the model loading codes
  • Only required models in a sequence would read data from .json files
  • Provide an extendable framework to integrate gitlib2 in the future.

Was presented at 06 Aug WP2 meeting. Also adding documentation and examples of how new model can be created to Allen docs.

Example of usage:

#include "MultiLayerFCNN.cuh"
using NeuralNetworkType = Allen::MVAModels::MultiLayerNN;
NeuralNetworkType two_track_mva_nn {"two_track_mva_nn", "/allen_two_track_mva_model_June22.json"};

All existing MVA models are migrated to the framework in MR: SingleLayerFCNN (Matching Ghost Killer, Downstream Selector etc), MultiLayerFCNN (Two Track MVA, ElectronID NN), CatboostDesTree (MuonCatboostEvaluator).

FYI: @adeoyang, @dovombru, @mveghel

Edited by Valerii Kholoimov

Merge request reports