Models
Machine learning model implementations for physics analysis.
Overview
The models module contains neural network architectures:
RecursiveGraphNeuralNetwork: Recursive message-passing GNN
grift: Graph Inference with Feature Transformation
Model Architecture
Recursive Graph Neural Network
A recursive message-passing architecture designed for physics graphs:
Multiple message-passing rounds
Node and edge updates
Graph-level readout
Customizable aggregation
GRIFT Model
Graph Inference with Feature Transformation model for learning on physics graphs.
Model Integration
Models in AnalysisG:
Inherit from
model_templateDefine network architecture in C++/Python
Implement forward pass
Specify loss functions and metrics
Configure optimizers
Training Workflow
The model template handles:
Data Loading: Batching and shuffling
Training Loop: Forward/backward passes
Validation: Periodic evaluation
Checkpointing: Saving best models
Logging: Metrics and losses
Early Stopping: Based on validation performance
Models can be trained using:
Standard supervised learning
Semi-supervised learning
Self-supervised learning
Transfer learning
Inference
Trained models can be used for:
Batch inference on new data
Real-time event classification
Feature extraction
Uncertainty quantification