57544b1ff9f97d4da9f64d25c8ea5a3d8d247ffc,rllib/examples/rock_paper_scissors_multiagent.py,,run_heuristic_vs_learned,#Any#Any#Any#,106

Before Change

            "policy_mapping_fn": select_policy,
        },
    }
    tune.run(trainer, stop={"timesteps_total": args.stop}, config=config)


def run_with_custom_entropy_loss(args):
    Example of customizing the loss function of an existing policy.

After Change

    trainer_obj = cls(config=config)
    env = trainer_obj.workers.local_worker().env
    for _ in range(args.stop_iters):
        results = trainer_obj.train()
        print(results)
        // Timesteps reached.
        if results["timesteps_total"] > args.stop_timesteps:
            break
        // Reward (difference) reached -> all good, return.
        elif env.player1_score - env.player2_score > args.stop_reward:
            return

    // Reward (difference) not reached: Error if `as_test`.
    if args.as_test:
        raise ValueError(
            "Desired reward difference ({}) not reached! Only got to {}.".
            format(args.stop_reward, env.player1_score - env.player2_score))

Instances