Function runSeededEpisode

runSeededEpisode<State, Command, Event>(
    environment: TrainingEnvironment<State, Command, Event>,
    choose: (
        observations: readonly NeuralObservation[],
    ) => readonly (number | null)[],
    options: { seed: number },
): SeededRun
Plays one whole run with the AI choosing, from a named seed: the same seed and the same choose make the same decisions, so a run is comparable between builds, policies and machines. choose must be deterministic for that to hold - the deterministic argmax of a NeuralPolicy, a heuristic's fixed candidate order - and receives the env's full roster, returning one action per agent. Unlike runRollouts, the chooser is not a NeuralModel: a heuristic policy, a trained model behind NeuralPolicy.selectAction, or any function a benchmark can name works the same way. The env's own maxSteps is the truncation bound.
Type Parameters
- State
- Command
- Event
Parameters
- environment: TrainingEnvironment<State, Command, Event>
- choose: (observations: readonly NeuralObservation[]) => readonly (number | null)[]
- options: { seed: number }
Returns SeededRun
Example
```
import { TrainingEnvironment, runSeededEpisode } from '@datamoc/mwg/simulation';
declare const env: TrainingEnvironment<{ x: number }, number, never>;
const run = runSeededEpisode(env, (observations) => [0], { seed: 7 });
console.log(run.steps, run.terminated);
```
- Defined in src/simulation/Training.ts:499