class SnakeEnv(Env):
def __init__(self):
self.action_space = Discrete(3) # 0 = turn left, 1 = do nothing, 2 = turn right
self.state = [0, 0, 1, 0]
self.game = Game()
self.reward = 0
self.done = False
def step(self, action):
offset = (action - 1)
translated_action = offset + self.game.snake.direction
if translated_action < 0:
translated_action = 3
if translated_action > 3:
translated_action = 0
self.reward, self.done = self.game.run(1, translated_action)
diff = (self.game.food.position[0] - self.game.snake.snake[0][0], self.game.food.position[1] - self.game.snake.snake[0][1])
self.state[0] = int(diff[0] < 0)
self.state[2] = int(diff[0] > 1)
self.state[1] = int(diff[1] < 0)
self.state[3] = int(diff[1] > 0)
return self.state, self.reward, self.done, {}
def render(self):
self.game.render()
def reset(self):
self.game.reset()
class SnakeWrapper:
"""
return the croped square_size-by-square_size after rotation and changing to one-hot and doing block-notation.
"""
# num_classes is the number of different element types that can be found on the board.
# yes I know, actually we have 9 types, but 10 is nicer. (4 snakes + 1 obstacle + 3 fruits + 1 empty = 9)
num_classes = 10
# the action space. 0-left, 1-forward, 2-right.
action_space = gym.spaces.Discrete(3)
# the observation space. 9x9 one hot vectors, total 9x9x10.
# your snake always look up (the observation is a rotated crop of the board).
observation_space = gym.spaces.Box(
low=0,
high=num_classes,
shape=(9, 9, 10),
dtype=np.int
)
def __init__(self):
self.game = Game()
self.square_size = 9 # the observation size
self.timestep = 0
def step(self, action):
# get action as integer, move the game one step forward
# return tuple: state, reward, done, info. done is always False - Snake game never ends.
action = int_to_action[action]
reward = self.game.step(action)
head_pos = self.game.players[1].chain[-1]
direction = self.game.players[1].direction
board = self.game.board
state = preprocess_snake_state(board, head_pos, direction, self.square_size, SnakeWrapper.num_classes)
self.timestep += 1
return state, reward
def seed(self, seed=None):
return self.game.seed(seed)
# reset the game and return the board observation
def reset(self):
self.game.reset()
self.timestep = 0
first_state, _ = self.step(0)
return first_state
# print the board to the console
def render(self, mode='human'):
self.game.render(self.timestep)
class SnakeEnv(Environment):
""" A (terribly simplified) Blackjack game implementation of an environment. """
def __init__(self, indim, outdim):
super().__init__()
""" All tasks are coupled to an environment. """
# the number of action values the environment accepts
self.indim = indim
# the number of sensor values the environment produces
self.outdim = outdim
self.game = None
self.running = True
self.numActions = 4
self.allActions = [
pygame.K_UP, pygame.K_DOWN, pygame.K_RIGHT, pygame.K_LEFT
]
self.stochAction = 0.
self.apple_distance = 0.
self.apple_change = 0.
def init_game(self, snake_size):
self.game = Game()
self.game.init_game(snake_size)
self.running = True
def getSensors(self):
""" the currently visible state of the world (the observation may be stochastic - repeated calls returning different values)
:rtype: by default, this is assumed to be a numpy array of doubles
"""
self.apple_distance = self.game.get_apple_distance()
state = self.game.get_current_state()
print(state)
index = 9 * state["left"] + 3 * state["forward"] + state["right"]
print(index)
return [
float(index),
]
def performAction(self, action):
""" perform an action on the world that changes it's internal state (maybe stochastically).
:key action: an action that should be executed in the Environment.
:type action: by default, this is assumed to be a numpy array of doubles
"""
action = int(action[0])
if self.stochAction > 0:
if random() < self.stochAction:
print(random())
action = choice(list(range(len(self.allActions))))
keydown = self.allActions[action]
self.game.update_frame(keydown)
if self.game.info["done"]:
self.running = False
return self.running
self.apple_change = self.apple_distance - self.game.get_apple_distance(
)
self.game.render()
if action == 0:
print("up")
if action == 1:
print("down")
if action == 2:
print("right")
if action == 3:
print("left")
def reset(self):
""" Most environments will implement this optional method that allows for reinitialization.