How to use the magent.ProcessingModel function in magent

def play_wrapper(model_names, n_rounds):
        time_stamp = time.time()

        models = []
        for i, item in enumerate(model_names):
            models.append(magent.ProcessingModel(env, handles[i], item[1], 0, item[-1]))

        for i, item in enumerate(model_names):
            models[i].load(item[0], item[2])

        leftID, rightID = 0, 1
        result = 0
        total_num = np.zeros(2)
        for _ in range(n_rounds):
            round_num = play(env, handles, models, map_size, leftID, rightID)
            total_num += round_num
            leftID, rightID = rightID, leftID
            result += 1 if round_num[0] > round_num[1] else 0
        result = 1.0 * result

        for model in models:
            model.quit()

base_args = {'batch_size': batch_size / unroll_step, 'unroll_step': unroll_step,
                     'memory_size': 8 * 300,
                     'target_update': target_update, 'train_freq': train_freq}
    elif args.alg == 'a2c':
        raise NotImplementedError
    else:
        raise NotImplementedError

    # load models
    names = [args.name + "-l0", args.name + "-l1", args.name + "-r0", args.name + "-r1"]
    models = []

    for i in range(len(names)):
        model_args = {'eval_obs': eval_obs[i]}
        model_args.update(base_args)
        models.append(magent.ProcessingModel(env, handles[i], names[i], 20000+i, 1000, RLModel, **model_args))

    # load if
    savedir = 'save_model'
    if args.load_from is not None:
        start_from = args.load_from
        print("load ... %d" % start_from)
        for model in models:
            model.load(savedir, start_from)
    else:
        start_from = 0

    # print state info
    print(args)
    print("view_size", env.get_view_space(handles[0]))
    print("feature_size", env.get_feature_space(handles[0]))

train_freq = 5

    models = []

    # load opponent
    if args.opponent >= 0:
        from magent.builtin.tf_model import DeepQNetwork
        models.append(magent.ProcessingModel(env, handles[1], names[1], 20000, 0, DeepQNetwork))
        models[0].load("data/battle_model", args.opponent)
    else:
        models.append(magent.ProcessingModel(env, handles[1], names[1], 20000, 0, RandomActor))

    # load our model
    if args.alg == 'dqn':
        from magent.builtin.tf_model import DeepQNetwork
        models.append(magent.ProcessingModel(env, handles[0], names[0], 20001, 1000, DeepQNetwork,
                                   batch_size=batch_size,
                                   learning_rate=3e-4,
                                   memory_size=2 ** 20, train_freq=train_freq, eval_obs=eval_obs[0]))
                                   
        step_batch_size = None
    elif args.alg == 'drqn':
        from magent.builtin.tf_model import DeepRecurrentQNetwork
        models.append(magent.ProcessingModel(env, handles[0], names[0], 20001, 1000, DeepRecurrentQNetwork,
                                   batch_size=batch_size/unroll_step, unroll_step=unroll_step,
                                   learning_rate=3e-4,
                                   memory_size=4 * 625, train_freq=train_freq, eval_obs=eval_obs[0]))
        step_batch_size = None
    elif args.alg == 'a2c':
        from magent.builtin.mx_model import AdvantageActorCritic
        step_batch_size = 10 * args.map_size * args.map_size * 0.04
        models.append(magent.ProcessingModel(env, handles[0], names[0], 20001, 1000, AdvantageActorCritic,

base_args = {'batch_size': batch_size / unroll_step, 'unroll_step': unroll_step,
                     'memory_size': 8 * 625, 'learning_rate': 1e-4,
                     'target_update': target_update, 'train_freq': train_freq}
    elif args.alg == 'a2c':
        raise NotImplementedError
    else:
        raise NotImplementedError

    # init models
    names = [args.name + "-l", args.name + "-r"]
    models = []

    for i in range(len(names)):
        model_args = {'eval_obs': eval_obs[i]}
        model_args.update(base_args)
        models.append(magent.ProcessingModel(env, handles[i], names[i], 20000, 1000, RLModel, **model_args))

    # load if
    savedir = 'save_model'
    if args.load_from is not None:
        start_from = args.load_from
        print("load ... %d" % start_from)
        for model in models:
            model.load(savedir, start_from)
    else:
        start_from = 0

    # print state info
    print(args)
    print("view_space", env.get_view_space(handles[0]))
    print("feature_space", env.get_feature_space(handles[0]))

batch_size=batch_size,
                                   learning_rate=3e-4,
                                   memory_size=2 ** 20, train_freq=train_freq, eval_obs=eval_obs[0]))
                                   
        step_batch_size = None
    elif args.alg == 'drqn':
        from magent.builtin.tf_model import DeepRecurrentQNetwork
        models.append(magent.ProcessingModel(env, handles[0], names[0], 20001, 1000, DeepRecurrentQNetwork,
                                   batch_size=batch_size/unroll_step, unroll_step=unroll_step,
                                   learning_rate=3e-4,
                                   memory_size=4 * 625, train_freq=train_freq, eval_obs=eval_obs[0]))
        step_batch_size = None
    elif args.alg == 'a2c':
        from magent.builtin.mx_model import AdvantageActorCritic
        step_batch_size = 10 * args.map_size * args.map_size * 0.04
        models.append(magent.ProcessingModel(env, handles[0], names[0], 20001, 1000, AdvantageActorCritic,
                                             learning_rate=1e-3))

    # load if
    savedir = 'save_model'
    if args.load_from is not None:
        start_from = args.load_from
        print("load ... %d" % start_from)
        models[0].load(savedir, start_from)
    else:
        start_from = 0

    # print debug info
    print(args)
    print("view_size", env.get_view_space(handles[0]))
    print("feature_size", env.get_feature_space(handles[0]))

RLModel = DeepRecurrentQNetwork
        base_args = {'batch_size': batch_size / unroll_step, 'unroll_step': unroll_step,
                     'memory_size': 8 * 625, 'learning_rate': 1e-4,
                     'target_update': target_update, 'train_freq': train_freq}
    elif args.alg == 'a2c':
        # see train_against.py to know how to use a2c
        raise NotImplementedError

    # init models
    names = [args.name + "-l", args.name + "-r"]
    models = []

    for i in range(len(names)):
        model_args = {'eval_obs': eval_obs[i]}
        model_args.update(base_args)
        models.append(magent.ProcessingModel(env, handles[i], names[i], 20000+i, 1000, RLModel, **model_args))

    # load if
    savedir = 'save_model'
    if args.load_from is not None:
        start_from = args.load_from
        print("load ... %d" % start_from)
        for model in models:
            model.load(savedir, start_from)
    else:
        start_from = 0

    # print state info
    print(args)
    print("view_space", env.get_view_space(handles[0]))
    print("feature_space", env.get_feature_space(handles[0]))