Python ALEInterface.reset_game方法代码示例

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class pyrlcade_environment(object):
    def init(self,rom_file,ale_frame_skip):

        self.ale = ALEInterface()

        self.max_frames_per_episode = self.ale.getInt("max_num_frames_per_episode");
        self.ale.set("random_seed",123)
        self.ale.set("disable_color_averaging",1)
        self.ale.set("frame_skip",ale_frame_skip)

        self.ale.loadROM(rom_file)
        self.legal_actions = self.ale.getMinimalActionSet()
        ram_size = self.ale.getRAMSize()
        self.ram = np.zeros((ram_size),dtype=np.uint8)
        self.ale.getRAM(self.ram)

        self.state = self.ale.getRAM(self.ram)

    def reset_state(self):
        self.ale.reset_game()

    def set_action(self,a):
        self.action = a

    def step(self):
        self.reward = self.ale.act(self.action)
        is_terminal = self.ale.game_over()
        return is_terminal

    def get_state(self):
        self.ale.getRAM(self.ram)
        return self.ram

    def get_reward(self):
        return self.reward

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Emulator(object):
    FRAME_SKIP = 4
    SCREEN_WIDTH = 84
    SCREEN_HEIGHT = 84

    def __init__(self, rom):
        self.ale = ALEInterface()
        self.max_num_frames_per_episode = 100000 #self.ale.getInt('max_num_frames_per_episode')
        self.ale.setInt('frame_skip', self.FRAME_SKIP)
        self.ale.loadROM('roms/' + rom)
        self.actions = self.ale.getMinimalActionSet()
        
    def reset(self):
        self.ale.reset_game()

    def image(self):
        screen = self.ale.getScreenGrayscale()
        screen = cv2.resize(screen, (self.SCREEN_HEIGHT, self.SCREEN_WIDTH))
        return np.reshape(screen, (self.SCREEN_HEIGHT, self.SCREEN_WIDTH))

    def act(self, action):
        return self.ale.act(action)

    def terminal(self):
        return self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Environment:
  def __init__(self, rom_file, args):
    self.ale = ALEInterface()
    if args.display_screen:
      if sys.platform == 'darwin':
        import pygame
        pygame.init()
        self.ale.setBool('sound', False) # Sound doesn't work on OSX
      elif sys.platform.startswith('linux'):
        self.ale.setBool('sound', True)
      self.ale.setBool('display_screen', True)

    self.ale.setInt('frame_skip', args.frame_skip)
    self.ale.setFloat('repeat_action_probability', args.repeat_action_probability)
    self.ale.setBool('color_averaging', args.color_averaging)

    if args.random_seed:
      self.ale.setInt('random_seed', args.random_seed)

    if args.record_screen_path:
      if not os.path.exists(args.record_screen_path):
        logger.info("Creating folder %s" % args.record_screen_path)
        os.makedirs(args.record_screen_path)
      logger.info("Recording screens to %s", args.record_screen_path)
      self.ale.setString('record_screen_dir', args.record_screen_path)

    if args.record_sound_filename:
      logger.info("Recording sound to %s", args.record_sound_filename)
      self.ale.setBool('sound', True)
      self.ale.setString('record_sound_filename', args.record_sound_filename)

    self.ale.loadROM(rom_file)

    if args.minimal_action_set:
      self.actions = self.ale.getMinimalActionSet()
      logger.info("Using minimal action set with size %d" % len(self.actions))
    else:
      self.actions = self.ale.getLegalActionSet()
      logger.info("Using full action set with size %d" % len(self.actions))
    logger.debug("Actions: " + str(self.actions))

    self.dims = (args.screen_height, args.screen_width)

  def numActions(self):
    return len(self.actions)

  def restart(self):
    self.ale.reset_game()

  def act(self, action):
    reward = self.ale.act(self.actions[action])
    return reward

  def getScreen(self):
    screen = self.ale.getScreenGrayscale()
    resized = cv2.resize(screen, self.dims)
    return resized

  def isTerminal(self):
    return self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Emulate:
  def __init__(self, rom_file, display_screen=False,frame_skip=4,screen_height=84,screen_width=84,repeat_action_probability=0,color_averaging=True,random_seed=0,record_screen_path='screen_pics',record_sound_filename=None,minimal_action_set=True):
    self.ale = ALEInterface()
    if display_screen:
      if sys.platform == 'darwin':
        import pygame
        pygame.init()
        self.ale.setBool('sound', False) # Sound doesn't work on OSX
      elif sys.platform.startswith('linux'):
        self.ale.setBool('sound', True)
      self.ale.setBool('display_screen', True)

    self.ale.setInt('frame_skip', frame_skip)
    self.ale.setFloat('repeat_action_probability', repeat_action_probability)
    self.ale.setBool('color_averaging', color_averaging)

    if random_seed:
      self.ale.setInt('random_seed', random_seed)

    self.ale.loadROM(rom_file)

    if minimal_action_set:
      self.actions = self.ale.getMinimalActionSet()
    else:
      self.actions = self.ale.getLegalActionSet()

    self.dims = (screen_width,screen_height)

  def numActions(self):
    return len(self.actions)

  def getActions(self):
  	return self.actions

  def restart(self):
    self.ale.reset_game()

  def act(self, action):
    reward = self.ale.act(self.actions[action])
    return reward

  def getScreen(self):
    screen = self.ale.getScreenGrayscale()
    resized = cv2.resize(screen, self.dims)
    return resized

  def getScreenGray(self):
    screen = self.ale.getScreenGrayscale()
    resized = cv2.resize(screen, self.dims)
    rotated = np.rot90(resized,k=1)
    return rotated

  def getScreenColor(self):
    screen = self.ale.getScreenRGB()
    resized = cv2.resize(screen, self.dims)
    rotated = np.rot90(resized,k=1)
    return rotated

  def isTerminal(self):
    return self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class AtariMDP(MDP, Serializable):

    def __init__(self, rom_path, obs_type=OBS_RAM, frame_skip=4):
        Serializable.__init__(self, rom_path, obs_type, frame_skip)
        self.options = (rom_path, obs_type, frame_skip)
        
        self.ale = ALEInterface()
        self.ale.loadROM(rom_path)        
        self._rom_path = rom_path
        self._obs_type = obs_type
        self._action_set = self.ale.getMinimalActionSet()
        self.frame_skip = frame_skip


    def get_image(self):
        return to_rgb(self.ale)
    def get_ram(self):
        return to_ram(self.ale)
    def game_over(self):
        return self.ale.game_over()
    def reset_game(self):
        return self.ale.reset_game()

    @property
    def n_actions(self):
        return len(self.action_set)

    def get_obs(self):
        if self._obs_type == OBS_RAM:
            return self.get_ram()[None,:]
        else:
            assert self._obs_type == OBS_IMAGE
            return self.get_image()[None,:,:,:]

    def step(self, a):

        reward = 0.0
        action = self.action_set[a]
        for _ in xrange(self.frame_skip):
            reward += self.ale.act(action)
        ob = self.get_obs().reshape(1,-1)
        return ob, np.array([reward]), self.ale.game_over()

    # return: (states, observations)
    def reset(self):
        self.ale.reset_game()
        return self.get_obs()

    @property
    def action_set(self):
        return self._action_set

    def plot(self):
        import cv2
        cv2.imshow("atarigame",self.get_image()) #pylint: disable=E1101
        cv2.waitKey(10) #pylint: disable=E1101

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Emulator:
    def __init__(self):
    
        self.ale = ALEInterface()
        
        # turn off the sound
        self.ale.setBool('sound', False)
        
        self.ale.setBool('display_screen', EMULATOR_DISPLAY)

        self.ale.setInt('frame_skip', FRAME_SKIP)
        self.ale.setFloat('repeat_action_probability', REPEAT_ACTION_PROBABILITY)
        self.ale.setBool('color_averaging', COLOR_AVERAGING)

        self.ale.setInt('random_seed', RANDOM_SEED)

        if RECORD_SCENE_PATH:
            self.ale.setString('record_screen_dir', RECORD_SCENE_PATH)


        self.ale.loadROM(ROM_PATH)

        self.actions = self.ale.getMinimalActionSet()
        logger.info("Actions: " + str(self.actions))

        self.dims = DIMS
        #self.start_lives = self.ale.lives()

    def getActions(self):
        return self.actions

    def numActions(self):
        return len(self.actions)

    def restart(self):
        self.ale.reset_game()
        # can be omitted

    def act(self, action):
        reward = self.ale.act(self.actions[action])
        return reward

    def getScreen(self):
        # why grayscale ?
        screen = self.ale.getScreenGrayscale()
        resized = cv2.resize(screen, self.dims)
        # normalize
        #resized /= COLOR_SCALE

        return resized

    def isTerminal(self):
        # while training deepmind only ends when agent dies
        #terminate = DEATH_END and TRAIN and (self.ale.lives() < self.start_lives)

        return self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Environment:
    def __init__(self, show_screen, history_length):
        self.ale = ALEInterface()
        self.ale.setInt('frame_skip', 4)
        self.history = None
        self.history_length = history_length
        if show_screen:
            self.display_screen()
        self.load_game()
        (screen_width, screen_height) = self.ale.getScreenDims()
        self.screen_data = np.empty((screen_height, screen_width, 1), dtype=np.uint8)  # 210x160 screen data
        self.dims = (84, 84)  # input size for neural network
        self.actions = [3, 0, 1, 4]  # noop, left, right, fire,

    def display_screen(self):
        self.ale.setBool("display_screen", True)

    def turn_on_sound(self):
        self.ale.setBool("sound", True)

    def restart(self):
        """reset game"""
        self.ale.reset_game()

    def act(self, action):
        """:returns reward of an action"""
        return self.ale.act(self.actions[action])

    def __get_screen(self):
        """:returns Grayscale thresholded resized screen image """
        self.ale.getScreenGrayscale(self.screen_data)
        resized = cv2.resize(self.screen_data, self.dims)
        return resized

    def get_state(self):
        binary_screen = self.__get_screen()
        if self.history is None:
            self.history = deque(maxlen=self.history_length)
            for _ in range(self.history_length - 1):
                self.history.append(binary_screen)
        self.history.append(binary_screen)
        result = np.stack(self.history, axis=0)
        return result

    def isTerminal(self):
        """checks if game is over"""
        return self.ale.game_over()

    def load_game(self):
        """load game from file"""
        self.ale.loadROM("Breakout.bin")

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Breakout(object):
    steps_between_actions = 4

    def __init__(self):
        self.ale = ALEInterface()
        self.ale.setInt('random_seed', 123)
        self.ale.setBool("display_screen", False)
        self.ale.setBool("sound", False)
        self.ale.loadROM("%s/breakout.bin" % rom_directory)
        self.current_state = [
            self.ale.getScreenRGB(), self.ale.getScreenRGB()
        ]

    def start_episode(self):
        self.ale.reset_game()

    def take_action(self, action):
        assert not self.terminated

        def step():
            reward = self.ale.act(action)
            self.roll_state()
            return reward

        reward = sum(step() for _ in xrange(self.steps_between_actions))

        return (reward, self.current_state)

    def roll_state(self):
        assert len(self.current_state) == 2
        self.current_state = [self.current_state[1], self.ale.getScreenRGB()]
        assert len(self.current_state) == 2

    @property
    def actions(self):
        return self.ale.getMinimalActionSet()

    @property
    def terminated(self):
        return self.ale.game_over() or self.ale.lives() < 5

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class emulator:
	def __init__(self, rom_name, vis):
		if vis:
			import cv2
		self.ale = ALEInterface()
		self.max_frames_per_episode = self.ale.getInt("max_num_frames_per_episode");
		self.ale.setInt("random_seed",123)
		self.ale.setInt("frame_skip",4)
		self.ale.loadROM('roms/' + rom_name )
		self.legal_actions = self.ale.getMinimalActionSet()
		self.action_map = dict()
		for i in range(len(self.legal_actions)):
			self.action_map[self.legal_actions[i]] = i

		# print(self.legal_actions)
		self.screen_width,self.screen_height = self.ale.getScreenDims()
		print("width/height: " +str(self.screen_width) + "/" + str(self.screen_height))
		self.vis = vis
		if vis: 
			cv2.startWindowThread()
			cv2.namedWindow("preview")

	def get_image(self):
		numpy_surface = np.zeros(self.screen_height*self.screen_width*3, dtype=np.uint8)
		self.ale.getScreenRGB(numpy_surface)
		image = np.reshape(numpy_surface, (self.screen_height, self.screen_width, 3))
		return image

	def newGame(self):
		self.ale.reset_game()
		return self.get_image()

	def next(self, action_indx):
		reward = self.ale.act(action_indx)	
		nextstate = self.get_image()
		# scipy.misc.imsave('test.png',nextstate)
		if self.vis:
			cv2.imshow('preview',nextstate)
		return nextstate, reward, self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Game():
    """
    Wrapper around the ALEInterface class.
    """

    def __init__(self, rom_file, sdl=False):
        self.ale = ALEInterface()
        # Setup SDL
        if sdl:
            if sys.platform == 'darwin':
                import pygame
                pygame.init()
                self.ale.setBool(b'sound', False) # Sound doesn't work on OSX
            elif sys.platform.startswith('linux'):
                self.ale.setBool(b'sound', True)
            self.ale.setBool(b'display_screen', True)

        # Load rom
        self.ale.loadROM(str.encode(rom_file))

    def get_action_set(self):
        return self.ale.getLegalActionSet()

    def get_minimal_action_set(self):
        return self.ale.getMinimalActionSet()

    def game_over(self):
        return self.ale.game_over()

    def act(self, action):
        return self.ale.act(action)

    def reset_game(self):
        self.ale.reset_game()

    def get_frame(self):
        return self.ale.getScreenRGB()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class AleInterface(object):
    def __init__(self, game, args):
        self.game = game
        self.ale = ALEInterface()

        # if sys.platform == 'darwin':
        #     self.ale.setBool('sound', False)  # Sound doesn't work on OSX
        # elif sys.platform.startswith('linux'):
        #     self.ale.setBool('sound', True)
        # self.ale.setBool('display_screen', True)
        #
        self.ale.setBool('display_screen', args.display_screen)

        self.ale.setInt('frame_skip', args.frame_skip)
        self.ale.setFloat('repeat_action_probability', args.repeat_action_probability)
        self.ale.setBool('color_averaging', args.color_averaging)
        self.ale.setInt('random_seed', args.random_seed)

        #
        # if rand_seed is not None:
        #     self.ale.setInt('random_seed', rand_seed)

        rom_file = "./roms/%s.bin" % game
        if not os.path.exists(rom_file):
            print "not found rom file:", rom_file
            sys.exit(-1)
        self.ale.loadROM(rom_file)

        self.actions = self.ale.getMinimalActionSet()


    def get_actions_num(self):
        return len(self.actions)

    def act(self, action):
        reward = self.ale.act(self.actions[action])
        return reward

    def get_screen_gray(self):
        return self.ale.getScreenGrayscale()

    def get_screen_rgb(self):
        return self.ale.getScreenRGB()

    def game_over(self):
        return self.ale.game_over()

    def reset_game(self):
        return self.ale.reset_game()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class Atari:
	def __init__(self,rom_name):
		self.ale = ALEInterface()
		self.max_frames_per_episode = self.ale.getInt("max_num_frames_per_episode")
		self.ale.setInt("random_seed",123)
		self.ale.setInt("frame_skip",4)
		self.ale.loadROM('./' +rom_name)
		self.screen_width,self.screen_height = self.ale.getScreenDims()
		self.legal_actions = self.ale.getMinimalActionSet()
		self.action_map = dict()
		for i in range(len(self.legal_actions)):
			self.action_map[self.legal_actions[i]] = i
		#print len(self.legal_actions)
		self.windowname = rom_name
		#cv2.startWindowThread()
		#cv2.namedWindow(rom_name)

	def get_image(self):
		numpy_surface = np.zeros(self.screen_height*self.screen_width*3, dtype=np.uint8)
		self.ale.getScreenRGB(numpy_surface)
		image = np.reshape(numpy_surface, (self.screen_height, self.screen_width, 3))
		return image

	def newGame(self):
		self.ale.reset_game()
		return self.get_image()

	def next(self, action):
		reward = self.ale.act(self.legal_actions[np.argmax(action)])	
		nextstate = self.get_image()
		
		#cv2.imshow(self.windowname,nextstate)
		if self.ale.game_over():
			self.newGame()
		#print "reward %d" % reward 
		return nextstate, reward, self.ale.game_over()

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
def get_random_baseline(gamepath):
    ale = ALEInterface()
    ale.setInt('random_seed', 42)

    recordings_dir = './recordings/breakout/'

    USE_SDL = True
    if USE_SDL:
        if sys.platform == 'darwin':
            import pygame
            pygame.init()
            ale.setBool('sound', False) # Sound doesn't work on OSX
            #ale.setString("record_screen_dir", recordings_dir);
        elif sys.platform.startswith('linux'):
            ale.setBool('sound', True)
            ale.setBool('display_screen', True)

    # Load the ROM file
    ale.loadROM(gamepath)

    # Get the list of legal actions
    legal_actions = ale.getLegalActionSet()

    # Play 5 episodes
    rewards = []
    for episode in xrange(10):
        total_reward = 0
        while not ale.game_over():
            a = legal_actions[randrange(len(legal_actions))]
            reward = ale.act(a);
            total_reward += reward
        rewards.append(total_reward)
        #print 'Episode', episode, 'ended with score:', total_reward
        ale.reset_game()
    avg_reward = sum(rewards) / float(len(rewards))
    return avg_reward

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class AtariGame(Task):
    ''' RL task based on Arcade Game.
    '''

    def __init__(self, rom_path, num_frames=4, live=False, skip_frame=0, mode='normal'):
        self.ale = ALEInterface()
        if live:
            USE_SDL = True
            if USE_SDL:
                if sys.platform == 'darwin':
                    import pygame
                    pygame.init()
                    self.ale.setBool('sound', False) # Sound doesn't work on OSX
                elif sys.platform.startswith('linux'):
                    self.ale.setBool('sound', True)
            self.ale.setBool('display_screen', True)
        self.mode = mode
        self.live = live
        self.ale.loadROM(rom_path)
        self.num_frames = num_frames
        self.frames = []
        self.frame_id = 0
        self.cum_reward = 0
        self.skip_frame = skip_frame
        if mode == 'small':
            img = T.matrix('img')
            self.max_pool = theano.function([img], max_pool_2d(img, [4, 4]))
            self.img_shape = (16, 16)
        else:
            self.img_shape = (84, 84) # image shape according to DQN Nature paper.
        while len(self.frames) < 4:
            self.step(choice(self.valid_actions, 1)[0])
        self.reset()


    def copy(self):
        import dill as pickle
        return pickle.loads(pickle.dumps(self))


    def reset(self):
        self.ale.reset_game()
        self.frame_id = 0
        self.cum_reward = 0
        if self.skip_frame:
            for frame_i in range(self.skip_frame):
                self.step(choice(self.valid_actions, 1)[0])


    @property
    def _curr_frame(self):
        img = self.ale.getScreenRGB()
        img = rgb2yuv(img)[:, :, 0] # get Y channel, according to Nature paper.
        # print 'RAM', self.ale.getRAM()
        if self.mode == 'small':
            img = self.max_pool(img)
        img = imresize(img, self.img_shape, interp='bicubic')
        return  img


    @property
    def curr_state(self):
        '''
        return raw pixels.
        '''
        return np.array(self.frames, dtype=floatX) / floatX(255.) # normalize


    @property
    def state_shape(self):
        return self.curr_state.shape


    @property
    def num_actions(self):
        return len(self.valid_actions)


    @property
    def valid_actions(self):
        return self.ale.getLegalActionSet()


    def step(self, action):
        reward = self.ale.act(action)
        if len(self.frames) == self.num_frames:
            self.frames = self.frames[1:]
        self.frames.append(self._curr_frame)
        self.frame_id += 1
        #print 'frame_id', self.frame_id
        self.cum_reward += reward
        return reward # TODO: scale the gradient up.


    def is_end(self):
        if np.abs(self.cum_reward) > 0:
            return True
        return self.ale.game_over()


#.........这里部分代码省略.........

# 需要导入模块: from ale_python_interface import ALEInterface [as 别名]
# 或者: from ale_python_interface.ALEInterface import reset_game [as 别名]
class AtariEnvironment:
    
    def __init__(self, args, outputDir):
        
        self.outputDir = outputDir
        self.screenCaptureFrequency = args.screen_capture_freq
        
        self.ale = ALEInterface()
        self.ale.setInt(b'random_seed', 123456)
        random.seed(123456)
        # Fix https://groups.google.com/forum/#!topic/deep-q-learning/p4FAIaabwlo
        self.ale.setFloat(b'repeat_action_probability', 0.0)

        # Load the ROM file
        self.ale.loadROM(args.rom)

        self.actionSet = self.ale.getMinimalActionSet()
        self.gameNumber = 0
        self.stepNumber = 0
        self.resetGame()

    def getNumActions(self):
        return len(self.actionSet)

    def getState(self):
        return self.state
    
    def getGameNumber(self):
        return self.gameNumber
    
    def getFrameNumber(self):
        return self.ale.getFrameNumber()
    
    def getEpisodeFrameNumber(self):
        return self.ale.getEpisodeFrameNumber()
    
    def getEpisodeStepNumber(self):
        return self.episodeStepNumber
    
    def getStepNumber(self):
        return self.stepNumber
    
    def getGameScore(self):
        return self.gameScore

    def isGameOver(self):
        return self.ale.game_over()

    def step(self, action):
        previousLives = self.ale.lives()
        reward = 0
        isTerminal = 0
        self.stepNumber += 1
        self.episodeStepNumber += 1
        
        for i in range(4):
            prevScreenRGB = self.ale.getScreenRGB()
            reward += self.ale.act(self.actionSet[action])
            screenRGB = self.ale.getScreenRGB()
    
            # Detect end of episode, I don't think I'm handling this right in terms
            # of the overall game loop (??)
            if self.ale.lives() < previousLives or self.ale.game_over():
                isTerminal = 1
                break

            if self.gameNumber % self.screenCaptureFrequency == 0:
                dir = self.outputDir + '/screen_cap/game-%06d' % (self.gameNumber)
                if not os.path.isdir(dir):
                    os.makedirs(dir)
                self.ale.saveScreenPNG(dir + '/frame-%06d.png' % (self.getEpisodeFrameNumber()))


        maxedScreen = np.maximum(screenRGB, prevScreenRGB)
        self.state = self.state.stateByAddingScreen(maxedScreen, self.ale.getFrameNumber())
        self.gameScore += reward
        return reward, self.state, isTerminal

    def resetGame(self):
        if self.ale.game_over():
            self.gameNumber += 1
        self.ale.reset_game()
        self.state = State().stateByAddingScreen(self.ale.getScreenRGB(), self.ale.getFrameNumber())
        self.gameScore = 0
        self.episodeStepNumber = 0 # environment steps vs ALE frames.  Will probably be 4*frame number