How-to¶
Recipes for common tasks. Each page assumes you already know what you want to do and just need the few-line answer.
| Task | Page |
|---|---|
| Drive the env from Gymnasium | Use the Gymnasium adapter |
| Drive the env from PettingZoo | Use the PettingZoo adapter |
| Plug into a POMDP solver | Use the POMDP adapter |
| Replace the opponent | Customize the opponent |
| Train the bandit instead of the guard | Switch the controlled side |
| Persist a rollout | Save and load runs |
| Run many seeds in parallel | vmap rollouts over seeds |
| Persist a scenario configuration | Round-trip config to JSON |
| Choose an observation pipeline | Pick an observation channel |
| Integrate an external solver | Plug in a solver |
See also¶
For deeper customization (writing your own reward / termination / IC sampler / observation function, plus pattern cookbooks for controlled and heuristic policies), see Extending.