def simulate_policy()

in scripts/run_goal_conditioned_policy.py [0:0]

• 28 lines of code
• 11 McCabe index (conditional complexity)

def simulate_policy(args):
    data = torch.load(args.file)
    policy = data['evaluation/policy']
    env = data['evaluation/env']
    print("Policy and environment loaded")
    if args.gpu:
        ptu.set_gpu_mode(True)
        policy.to(ptu.device)
    if isinstance(env, VAEWrappedEnv) and hasattr(env, 'mode'):
        env.mode(args.mode)
    if args.enable_render or hasattr(env, 'enable_render'):
        # some environments need to be reconfigured for visualization
        env.enable_render()
    paths = []
    while True:
        paths.append(multitask_rollout(
            env,
            policy,
            max_path_length=args.H,
            render=not args.hide,
            observation_key='observation',
            desired_goal_key='desired_goal',
        ))
        if hasattr(env, "log_diagnostics"):
            env.log_diagnostics(paths)
        if hasattr(env, "get_diagnostics"):
            for k, v in env.get_diagnostics(paths).items():
                logger.record_tabular(k, v)
        logger.dump_tabular()