Extensibility

This framework was designed to be extensible. Common extension points:

1. New compartment models

Add a new block in the YAML with compartments and transitions — no code changes needed.

2. New parameter types

Add parameters to the parameters map. For time-varying parameters, implement extras_fn.

3. Custom loss functions

Provide a Python callable to compute the loss instead of MSE. For example, weighted MSE or Poisson log-likelihood for count data.

4. Metapopulation / Network models

• Extend transitions syntax to include indices or use multiple population blocks.
• The ODE generator will need to be extended to create vectorized compartments per patch.

5. New optimizers or inference engines

• Add wrappers for other optimizers (e.g., CMA-ES, differential evolution).
• Add MCMC engines (e.g., PyMC, NumPyro) by mapping log-posterior.