How to use the tensorforce.agents.Agent.from_spec function in Tensorforce

path = os.path.dirname(__file__)
    if args.agent_config:
        # Use absolute path
        agent_config = json.load(path + args.agent_config)
    else:
        raise TensorForceError("No agent configuration provided.")

    if not args.network_spec:
        raise TensorForceError("No network configuration provided.")
    else:
        network_spec = json.load(path + args.network_config)

    logger = logging.getLogger(__name__)
    logger.setLevel(logging.INFO)  # configurable!!!

    agent = Agent.from_spec(
        spec=agent_config,
        kwargs=dict(
            states=environment.states,
            actions=environment.actions,
            network=network_spec
        )
    )
    if args.load:
        load_dir = os.path.dirname(args.load)
        if not os.path.isdir(load_dir):
            raise OSError("Could not load agent from {}: No such directory.".format(load_dir))
        agent.restore_model(args.load)

    if args.debug:
        logger.info("-" * 16)
        logger.info("Configuration:")

distributed_spec=dict(
            cluster_spec=cluster_spec,
            task_index=args.task_index,
            parameter_server=args.parameter_server,
            device=('/job:{}/task:{}'.format('ps' if args.parameter_server else 'worker', args.task_index)),  # '/cpu:0'
        )
    )

    if args.network_spec:
        with open(args.network_spec, 'r') as fp:
            network_spec = json.load(fp=fp)
    else:
        network_spec = None
        logger.info("No network configuration provided.")

    agent = Agent.from_spec(
        spec=args.agent,
        kwargs=dict(
            states_spec=environment.states,
            actions_spec=environment.actions,
            network_spec=network_spec,
            config=config
        )
    )

    logger.info("Starting distributed agent for OpenAI Gym '{gym_id}'".format(gym_id=args.gym_id))
    logger.info("Config:")
    logger.info(config)

    runner = Runner(
        agent=agent,
        environment=environment,

if args.network_spec is not None:
        with open(args.network_spec, 'r') as fp:
            network_spec = json.load(fp=fp)
    else:
        network_spec = None
        logger.info("No network configuration provided.")

    
    agent_config['states_preprocessing'] = [{'type': 'flatten'}]

    logger.info("Start training")

    environment = OpenSim(env_id=1,visualize=True)

    agent = Agent.from_spec(
        spec=agent_config,
        kwargs=dict(
            states=environment.states,
            actions=environment.actions,
            network=network_spec,
        )
    )

    runner = Runner(
        agent=agent,
        environment=environment,
        repeat_actions=1
    )

    def episode_finished(r):
        if r.episode % 100 == 0:

environment = MazeExplorer(mode_id=args.mode, visible=not args.hide)

    if args.agent_config is not None:
        with open(args.agent_config, 'r') as fp:
            agent_config = json.load(fp=fp)
    else:
        raise TensorForceError("No agent configuration provided.")

    if args.network_spec is not None:
        with open(args.network_spec, 'r') as fp:
            network = json.load(fp=fp)
    else:
        network = None
        logger.info("No network configuration provided.")

    agent = Agent.from_spec(
        spec=agent_config,
        kwargs=dict(
            states=environment.states,
            actions=environment.actions,
            network=network
        )
    )

    if args.load:
        load_dir = os.path.dirname(args.load)
        if not os.path.isdir(load_dir):
            raise OSError("Could not load agent from {}: No such directory.".format(load_dir))
        agent.restore_model(args.load)

    if args.debug:
        logger.info("-" * 16)