graphnet-team/graphnet

arguments:
  backbone:
    ModelConfig:
      arguments:
        add_global_variables_after_pooling: false
        dynedge_layer_sizes: null
        features_subset: null
        global_pooling_schemes: [min, max, mean, sum]
        nb_inputs: 4
        nb_neighbours: 8
        post_processing_layer_sizes: null
        readout_layer_sizes: null
      class_name: DynEdge
  graph_definition:
    ModelConfig:
      arguments:
        columns: [0, 1, 2]
        detector:
          ModelConfig:
            arguments: {}
            class_name: Prometheus
        dtype: null
        nb_nearest_neighbours: 8
        node_definition:
          ModelConfig:
            arguments: {}
            class_name: NodesAsPulses
        input_feature_names: [sensor_pos_x, sensor_pos_y, sensor_pos_z, t]
      class_name: KNNGraph
  optimizer_class: '!class torch.optim.adam Adam'
  optimizer_kwargs: {eps: 0.001, lr: 0.001}
  scheduler_class: '!class graphnet.training.callbacks PiecewiseLinearLR'
  scheduler_config: {interval: step}
  scheduler_kwargs:
    factors: [0.01, 1, 0.01]
    milestones: [0, 20.0, 80]
  tasks:
  - ModelConfig:
      arguments:
        nb_outputs: 3 # number of classes
        prediction_labels: ['noise', 'muon', 'neutrino']
        hidden_size: 128
        loss_function:
          ModelConfig:
            arguments: {options: {1: 0, -1: 0, 13: 1, -13: 1, 12: 2, -12: 2, 14: 2, -14: 2, 16: 2, -16: 2}}
            class_name: CrossEntropyLoss
        loss_weight: null
        target_labels: dummy_pid
        transform_inference: '!lambda x: torch.nn.functional.softmax(x, dim=-1)'
        transform_prediction_and_target: null
        transform_support: null
        transform_target: null
      class_name: MulticlassClassificationTask
class_name: StandardModel