Python ExtraTimeLimit示例

示例#1

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()
    elif args["env_kind"] == "my_games":
        env = gym.make(args['env'])
        env = MaxAndSkipEnv(env, skip=4)
        env = WarpFrame(env, width=84, height=84)
        env = FrameStack(env, 4)

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#2

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if add_monitor:
        if not os.path.exists(args['save_dir']+'csv_file/'):
            os.makedirs(args['save_dir']+'csv_file/')
        env = Monitor(env, osp.join(args['save_dir'], 'csv_file/%.2i' % rank))
    return env

示例#3

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == "atari":
        env = gym.make(args["env"])
        assert "NoFrameskip" in env.spec.id
        # from self-supervised exploration via disagreement
        if args["stickyAtari"] == "true":
            env = StickyActionEnv(env)
        env._max_episode_steps = args["max_episode_steps"] * 4
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args["max_episode_steps"])
        if "Montezuma" in args["env"]:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
        if args["noisy_tv"] == "true":
            env = NoisyTVEnvWrapper(env)
        # assert env.action_space == spaces.Discrete(7)
    elif args["env_kind"] == "mario":
        env = make_mario_env()
        if args["noisy_tv"] == "true":
            env = NoisyTVEnvWrapperMario(env)
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == "robopong":
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), "%.2i" % rank))
    return env

示例#4

文件： run.py 项目： ramanans1/large-scale-curiosity

def make_env_all_params(rank, add_monitor, args, logdir):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()
    elif args["env_kind"] == "dm_suite":
        env = make_dm_suite(task=args["env"],
                            logdir=logdir,
                            to_record=args["to_record"])

    if add_monitor:
        env = TempMonitor(env)

    return env

示例#5

def make_env_all_params(rank, add_monitor, args, sleep_multiple=2):
    if args["env_kind"] == 'ObstacleTowerEnv':
        env = _make_obs_env(rank, add_monitor, args, sleep_multiple)
    elif args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
        if rank == 2:
            env = RenderWrapper(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if add_monitor:
        logdir = osp.join('summaries', args["exp_name"])
        logger.configure(logdir)
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#6

def make_env_all_params(rank, add_monitor, args):
    from baselines import logger
    if args["env_kind"] == 'deepmind':
        if args["depth_pred"]:
            env = gym.make(args['env'])
            env = DeepmindLabMaze(env, args["env"], args['nsteps_per_seg'], depth=True)
        else:
            env = gym.make(args['env'])
            #env = ProcessFrame84(env, crop=False)
            #env = FrameStack(env, 4)
            env = DeepmindLabMaze(env, args["env"], args['nsteps_per_seg'])
    elif args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#7

文件： run.py 项目： denncli/multi-step-curiosity-driven-learning

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if args["env_kind"] == 'atari' and add_monitor:
        #env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
        env = Monitor(env,
                      os.path.join(os.getcwd(), 'test_video'),
                      force=True,
                      video_callable=lambda episode_id: episode_id % 20 == 0)
        #env = Monitor(env, os.path.join(os.getcwd(), 'test_video'),video_callable=lambda episode_id: True )#,force=True)

    return env

示例#8

文件： run.py 项目： RobinKa/large-scale-curiosity

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()
    elif args["env_kind"] == "virtualbox":
        print("Make virtualbox")
        env = make_virtualbox()

    if add_monitor:
        logger_dir = "log" if logger.get_dir() is None else logger.get_dir()
        print("Adding monitor at dir", logger_dir)
        env = Monitor(env, osp.join(logger_dir, '%.2i' % rank))
    return env

示例#9

文件： rnn_run.py 项目： Ryusizz/Pred-error-based-RL

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'field':
        import gym_fieldedmove
        env = gym.make('FieldedMove-v0')
        # env = FrameStack(env, 4)
    elif args["env_kind"] == "ple":
        import gym_ple
        env = gym.make(args['env'])
        env._max_episode_steps = args['max_episode_steps']
        # env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#10

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if args["dyn_env"]:
            env = MakeEnvDynamic(env)
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#11

文件： run.py 项目： okokchoi/impatiencerl

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'robopong':
        if args["env"] == "pong":
            env = make_robo_pong()
        elif args["env"] == "hockey":
            env = make_robo_hockey()

    if add_monitor:
        #print(osp.join(logger.get_dir(), '%.2i' % rank + '.monitor.csv'))

        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
        """
        env = DummyVecEnv([lambda: env])
        
        env = VecVideoRecorder(env, directory = './vid',
                       record_video_trigger=lambda step: step == 0,
                       video_length= 100,)
        
        env.reset()
        """
        #env = wrappers.Monitor(env,'./vid/',force = True,write_upon_reset = True, video_callable=lambda episode: True)
        #print(osp.join(logger.get_dir()))
        #env = Monitor(env, osp.join(logger.get_dir()))
        #env = Monitor(env,  "./vid", video_callable=lambda episode_id: True,force=True)
    return env

示例#12

文件： record_unwrapped_video.py 项目： self-supervisor/Escaping-Stochastic-Traps-With-Aleatoric-Mapping-Agents

    def make_atari_env(self, args):
        """
        duplicated code hack due to
        relative import errors
        """

        env = gym.make(args["env"])
        assert "NoFrameskip" in env.spec.id
        # from self-supervised exploration via disagreement
        if args["stickyAtari"] == "true":
            env = StickyActionEnv(env)
        env._max_episode_steps = args["max_episode_steps"] * 4
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStackNoLazy(env, 4)
        env = ExtraTimeLimit(env, args["max_episode_steps"])
        if "Montezuma" in args["env"]:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
        if args["noisy_tv"] == "true":
            env = NoisyTVEnvWrapper(env)
        return env

示例#13

文件： play.py 项目： Baichenjia/VDM

def make_env_all_params(rank, args):
    env = None
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        if args["stickyAtari"]:               # 在智能体执行动作时增加随机性
            env._max_episode_steps = args['max_episode_steps'] * 4
            env = StickyActionEnv(env)
        else:
            env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)         # 每个动作连续执行4步
        env = ProcessFrame84(env, crop=False)    # 处理观测
        env = FrameStack(env, 4)                 # 将连续4帧叠加起来作为输入
        if not args["stickyAtari"]:
            env = ExtraTimeLimit(env, args['max_episode_steps'])   # 限制了一个周期的最大时间步.
        if 'Montezuma' in args['env']:           # 记录智能体的位置, 所在的房间, 已经访问的房间
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':            # 超级马里奥
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":      # 多智能体游戏, Multi-Pong
        env = make_multi_pong()
    return env

示例#14

文件： run.py 项目： speedcell4/exploration-by-disagreement

def make_env_all_params(rank, add_monitor, args):
    if args["env_kind"] == 'atari':
        env = gym.make(args['env'])
        assert 'NoFrameskip' in env.spec.id
        if args["stickyAtari"]:
            env._max_episode_steps = args['max_episode_steps'] * 4
            env = StickyActionEnv(env)
        else:
            env = NoopResetEnv(env, noop_max=args['noop_max'])
        env = MaxAndSkipEnv(env, skip=4)
        env = ProcessFrame84(env, crop=False)
        env = FrameStack(env, 4)
        if not args["stickyAtari"]:
            env = ExtraTimeLimit(env, args['max_episode_steps'])
        if 'Montezuma' in args['env']:
            env = MontezumaInfoWrapper(env)
        env = AddRandomStateToInfo(env)
    elif args["env_kind"] == 'mario':
        env = make_mario_env()
    elif args["env_kind"] == "retro_multi":
        env = make_multi_pong()
    elif args["env_kind"] == 'unity':
        env = make_unity_maze(args["env"],
                              seed=args["seed"],
                              rank=rank,
                              ext_coeff=args["ext_coeff"],
                              recordUnityVid=args['recordUnityVid'],
                              expID=args["unityExpID"],
                              startLoc=args["startLoc"],
                              door=args["door"],
                              tv=args["tv"],
                              testenv=args["testenv"],
                              logdir=logger.get_dir())

    if add_monitor:
        env = Monitor(env, osp.join(logger.get_dir(), '%.2i' % rank))
    return env

示例#15

文件： bank_heist_diagram.py 项目： self-supervisor/Escaping-Stochastic-Traps-With-Aleatoric-Mapping-Agents

    ActionLoggingWrapper,
)


def get_car_mask(frames, car_color=np.array([223, 183, 85])):
    mask = np.zeros(shape=frames[0].shape)
    for a_frame in frames:
        for i in range(a_frame.shape[0]):
            for j in range(a_frame.shape[1]):
                if np.array_equal(a_frame[i][j], car_color):
                    mask[i][j] += 1
    return mask


env = gym.make("BankHeistNoFrameskip-v4")
env = gym.wrappers.Monitor(env, "./video/", force=True)
env._max_episode_steps = 4000 * 4
env = MaxAndSkipEnv(env, skip=4)
env = ProcessFrame84(env, crop=False)
env = FrameStack(env, 4)
env = ExtraTimeLimit(env, 4000)
env = AddRandomStateToInfo(env)

obs = env.reset()

for _ in range(100):
    obs, reward, done, info = env.step(env.action_space.sample())
    import pdb

    pdb.set_trace()