Model training
This section is related to the JUNE-NZ component: cli_train. It is used to train mutiple model parmaters to fit the provided target data.
1. Background
At the core of this model lies a SIR (Susceptible -> Infected -> Recovered) procedure.
All individual agents within the model engage in interactions using a Graph Neural Network (GNN).
The GNN facilitates the application of various policies within the simulated society, including actions like school closures and vaccination campaigns.
The model’s behavior is governed by a range of parameters specified in the configuration file. These parameters are initially set using a Long Short-Term Memory Neural Network, and they are fine-tuned during the model’s iterations through backpropagation. This process involves minimizing the error between the model’s predictions and the desired outcomes.