def step_generator(self):
state = self
if self.cfs is None:
self.cfs = CFS(sensors=state.sensors().keys(),
actions=state.actions().keys())
for k in xrange(2000):
self.cfs.add_message(state.sensors_msg())
action = self.cfs.get_action()
try:
state = state.act(action)
except garlicsim.misc.WorldEnded:
self.cfs.clear_messages()
self.cfs.give_feedback(state.points)
self.cfs.evolve()
state = State()
yield state
raise garlicsim.misc.WorldEnded
def step_generator(self):
state = self
if self.cfs is None:
self.cfs = CFS(sensors=state.sensors().keys(),
actions=state.actions().keys())
for k in xrange(2000):
self.cfs.add_message(state.sensors_msg())
action = self.cfs.get_action()
try:
state = state.act(action)
except garlicsim.misc.WorldEnded:
self.cfs.clear_messages()
self.cfs.give_feedback(state.points)
self.cfs.evolve()
state = State()
yield state
raise garlicsim.misc.WorldEnded
class State(garlicsim.data_structures.State):
def __init__(self,
board=None,
player_pos=(0, 0),
points=0,
player_dir='up',
bump=False,
player_has_arrow=True,
wumpus_dead=False,
scream=False,
gold_grabbed=False,
game_won=False,
shooting=False,
cfs=None):
self.winning_prize = 1000
self.death_penalty = -1000
self.action_penalty = -1
self.arrow_penalty = -10
self.board_size = 4
self.player_pos = player_pos
self.player_dir = player_dir
self.points = points
self.board = board or self._initiate_board()
self.bump = bump
self.player_has_arrow = player_has_arrow
self.wumpus_dead = wumpus_dead
self.scream = scream
self.gold_grabbed = gold_grabbed
self.game_won = game_won
self.shooting = shooting
self.cfs = cfs
def _initiate_board(self):
board = {}
pit_probability = 0.2
is_pit = lambda: random() <= pit_probability
field_indices = sorted(list(product(xrange(self.board_size),
repeat=2)))
board[0, 0] = BoardTile(pit=False, wumpus=False, gold=False)
for indices in field_indices[1:]:
board[indices] = BoardTile(pit=is_pit(), wumpus=False, gold=False)
random_field_indices = lambda: choice(field_indices[1:])
wumpus = random_field_indices()
gold = random_field_indices()
board[wumpus] = board[wumpus]._replace(wumpus=True)
board[gold] = board[gold]._replace(gold=True)
return board
def next_state(self, *args, **kwargs):
if self.current_field().wumpus and not self.wumpus_dead:
raise garlicsim.misc.WorldEnded
if self.current_field().pit:
raise garlicsim.misc.WorldEnded
if self.game_won:
raise garlicsim.misc.WorldEnded
defaults = {
'board': self.board,
'player_pos': self.player_pos,
'player_dir': self.player_dir,
'points': self.points - 1,
'bump': False,
'player_has_arrow': self.player_has_arrow,
'wumpus_dead': self.wumpus_dead,
'scream': False,
'gold_grabbed': self.gold_grabbed,
'game_won': self.game_won,
'shooting': False,
'cfs': self.cfs
}
defaults.update(kwargs)
return State(**defaults)
def step_turn_left(self):
new_dir = {
'up': 'left',
'left': 'down',
'down': 'right',
'right': 'up'
}[self.player_dir]
return self.next_state(player_dir=new_dir)
def step_turn_right(self):
new_dir = {
'up': 'right',
'left': 'up',
'down': 'left',
'right': 'down'
}[self.player_dir]
return self.next_state(player_dir=new_dir)
def step_shoot_arrow(self):
if not self.player_has_arrow:
return self.next_state()
next_state = self.next_state(player_has_arrow=False)
next_state.points += self.arrow_penalty
x, y = self.player_pos
arrow_path = {
'up': [(x, k) for k in range(y, self.board_size)],
'left': [(k, y) for k in range(0, x)],
'down': [(x, k) for k in range(0, y)],
'right': [(k, y) for k in range(x, self.board_size)]
}[self.player_dir]
for pos in arrow_path:
if self.board[pos].wumpus:
next_state.wumpus_dead = True
next_state.scream = True
break
if self.player_has_arrow:
next_state.shooting = True
return next_state
def step_go_forward(self):
x, y = self.player_pos
new_pos = {
'up': lambda: (x, y + 1),
'left': lambda: (x - 1, y),
'down': lambda: (x, y - 1),
'right': lambda: (x + 1, y)
}[self.player_dir]()
new_x, new_y = new_pos
if (0 <= new_x < self.board_size) and (0 <= new_y < self.board_size):
next_state = self.next_state(player_pos=new_pos)
else:
next_state = self.next_state(bump=True)
if next_state.current_field().wumpus and not next_state.wumpus_dead:
next_state.points += self.death_penalty
if next_state.current_field().pit:
next_state.points += self.death_penalty
return next_state
def step_grab_gold(self):
if self.board[self.player_pos].gold:
return self.next_state(gold_grabbed=True)
else:
return self.next_state()
def step_climb_out(self):
if self.player_pos == (0, 0):
next_state = self.next_state(game_won=True)
if self.gold_grabbed:
next_state.points += self.winning_prize
else:
next_state = self.next_state()
return next_state
def sensors(self):
sensors = {
'breeze': False,
'glow': False,
'stink': False,
'scream': self.scream,
'bump': self.bump,
'gold': self.gold_grabbed,
'arrow': self.player_has_arrow
}
x, y = self.player_pos
adjacent_fields = [(x, y)]
adjacent_fields += [(k, y) for k in [x - 1, x + 1]]
adjacent_fields += [(x, k) for k in [y - 1, y + 1]]
for field in adjacent_fields:
if field in self.board:
tile = self.board[field]
if tile.wumpus and not self.wumpus_dead:
sensors['stink'] = True
if tile.pit:
sensors['breeze'] = True
if self.current_field().gold and not self.gold_grabbed:
sensors['glow'] = True
return sensors
def sensors_msg(self):
sensors = self.sensors()
message = "10" #input message
for key in sorted(sensors.keys()):
if sensors[key]:
message += '1'
else:
message += '0'
return message
def current_field(self):
return self.board[self.player_pos]
def actions(self):
actions = {
'go_up': lambda: self.go('up'),
'go_left': lambda: self.go('left'),
'go_down': lambda: self.go('down'),
'go_right': lambda: self.go('right'),
'shoot_up': lambda: self.shoot_to('up'),
'shoot_left': lambda: self.shoot_to('left'),
'shoot_down': lambda: self.shoot_to('down'),
'shoot_right': lambda: self.shoot_to('right'),
'grab': lambda: self.step_grab_gold(),
'climb': lambda: self.step_climb_out()
}
return actions
def act(self, action):
return self.actions()[action]()
def turn_to(self, direction):
directions = {'up': 0, 'left': 1, 'down': 2, 'right': 3}
curr_idx = directions[self.player_dir]
new_idx = directions[direction]
diff = (curr_idx - new_idx) % 4
if diff == 0:
return self
if diff == 1:
return self.step_turn_right()
if diff == 2:
return self.step_turn_right().step_turn_right()
if diff == 3:
return self.step_turn_left()
def go(self, direction):
return self.turn_to(direction).step_go_forward()
def shoot_to(self, direction):
return self.turn_to(direction).step_shoot_arrow()
def step_generator(self):
state = self
if self.cfs is None:
self.cfs = CFS(sensors=state.sensors().keys(),
actions=state.actions().keys())
for k in xrange(2000):
self.cfs.add_message(state.sensors_msg())
action = self.cfs.get_action()
try:
state = state.act(action)
except garlicsim.misc.WorldEnded:
self.cfs.clear_messages()
self.cfs.give_feedback(state.points)
self.cfs.evolve()
state = State()
yield state
raise garlicsim.misc.WorldEnded
@staticmethod
def create_root():
return State()
class State(garlicsim.data_structures.State):
def __init__(self, board=None, player_pos=(0, 0), points=0,
player_dir='up', bump=False, player_has_arrow=True,
wumpus_dead=False, scream=False, gold_grabbed=False,
game_won=False, shooting=False, cfs=None):
self.winning_prize = 1000
self.death_penalty = -1000
self.action_penalty = -1
self.arrow_penalty = -10
self.board_size = 4
self.player_pos = player_pos
self.player_dir = player_dir
self.points = points
self.board = board or self._initiate_board()
self.bump = bump
self.player_has_arrow = player_has_arrow
self.wumpus_dead = wumpus_dead
self.scream = scream
self.gold_grabbed = gold_grabbed
self.game_won = game_won
self.shooting = shooting
self.cfs = cfs
def _initiate_board(self):
board = {}
pit_probability = 0.2
is_pit = lambda: random() <= pit_probability
field_indices = sorted(list(product(xrange(self.board_size), repeat=2)))
board[0,0] = BoardTile(pit=False, wumpus=False, gold=False)
for indices in field_indices[1:]:
board[indices] = BoardTile(pit=is_pit(), wumpus=False, gold=False)
random_field_indices = lambda: choice(field_indices[1:])
wumpus = random_field_indices()
gold = random_field_indices()
board[wumpus] = board[wumpus]._replace(wumpus=True)
board[gold] = board[gold]._replace(gold=True)
return board
def next_state(self, *args, **kwargs):
if self.current_field().wumpus and not self.wumpus_dead:
raise garlicsim.misc.WorldEnded
if self.current_field().pit:
raise garlicsim.misc.WorldEnded
if self.game_won:
raise garlicsim.misc.WorldEnded
defaults = {
'board': self.board,
'player_pos': self.player_pos,
'player_dir': self.player_dir,
'points': self.points - 1,
'bump': False,
'player_has_arrow': self.player_has_arrow,
'wumpus_dead': self.wumpus_dead,
'scream': False,
'gold_grabbed': self.gold_grabbed,
'game_won': self.game_won,
'shooting': False,
'cfs': self.cfs
}
defaults.update(kwargs)
return State(**defaults)
def step_turn_left(self):
new_dir = {
'up': 'left',
'left': 'down',
'down': 'right',
'right': 'up'
}[self.player_dir]
return self.next_state(player_dir=new_dir)
def step_turn_right(self):
new_dir = {
'up': 'right',
'left': 'up',
'down': 'left',
'right': 'down'
}[self.player_dir]
return self.next_state(player_dir=new_dir)
def step_shoot_arrow(self):
if not self.player_has_arrow:
return self.next_state()
next_state = self.next_state(player_has_arrow=False)
next_state.points += self.arrow_penalty
x, y = self.player_pos
arrow_path = {
'up': [(x, k) for k in range(y, self.board_size)],
'left': [(k, y) for k in range(0, x)],
'down': [(x, k) for k in range(0, y)],
'right': [(k, y) for k in range(x, self.board_size)]
}[self.player_dir]
for pos in arrow_path:
if self.board[pos].wumpus:
next_state.wumpus_dead = True
next_state.scream = True
break
if self.player_has_arrow:
next_state.shooting = True
return next_state
def step_go_forward(self):
x, y = self.player_pos
new_pos = {
'up': lambda: (x, y + 1),
'left': lambda: (x - 1, y),
'down': lambda: (x, y - 1),
'right': lambda: (x + 1, y)
}[self.player_dir]()
new_x, new_y = new_pos
if (0 <= new_x < self.board_size) and (0 <= new_y < self.board_size):
next_state = self.next_state(player_pos=new_pos)
else:
next_state = self.next_state(bump=True)
if next_state.current_field().wumpus and not next_state.wumpus_dead:
next_state.points += self.death_penalty
if next_state.current_field().pit:
next_state.points += self.death_penalty
return next_state
def step_grab_gold(self):
if self.board[self.player_pos].gold:
return self.next_state(gold_grabbed=True)
else:
return self.next_state()
def step_climb_out(self):
if self.player_pos == (0, 0):
next_state = self.next_state(game_won=True)
if self.gold_grabbed:
next_state.points += self.winning_prize
else:
next_state = self.next_state()
return next_state
def sensors(self):
sensors = {
'breeze': False,
'glow': False,
'stink': False,
'scream': self.scream,
'bump': self.bump,
'gold': self.gold_grabbed,
'arrow': self.player_has_arrow
}
x, y = self.player_pos
adjacent_fields = [(x, y)]
adjacent_fields += [(k, y) for k in [x - 1, x + 1]]
adjacent_fields += [(x, k) for k in [y - 1, y + 1]]
for field in adjacent_fields:
if field in self.board:
tile = self.board[field]
if tile.wumpus and not self.wumpus_dead:
sensors['stink'] = True
if tile.pit:
sensors['breeze'] = True
if self.current_field().gold and not self.gold_grabbed:
sensors['glow'] = True
return sensors
def sensors_msg(self):
sensors = self.sensors()
message = "10" #input message
for key in sorted(sensors.keys()):
if sensors[key]:
message += '1'
else:
message += '0'
return message
def current_field(self):
return self.board[self.player_pos]
def actions(self):
actions = {
'go_up': lambda: self.go('up'),
'go_left': lambda: self.go('left'),
'go_down': lambda: self.go('down'),
'go_right': lambda: self.go('right'),
'shoot_up': lambda: self.shoot_to('up'),
'shoot_left': lambda: self.shoot_to('left'),
'shoot_down': lambda: self.shoot_to('down'),
'shoot_right': lambda: self.shoot_to('right'),
'grab': lambda: self.step_grab_gold(),
'climb': lambda: self.step_climb_out()
}
return actions
def act(self, action):
return self.actions()[action]()
def turn_to(self, direction):
directions = {'up': 0, 'left': 1, 'down': 2, 'right': 3}
curr_idx = directions[self.player_dir]
new_idx = directions[direction]
diff = (curr_idx - new_idx) % 4
if diff == 0:
return self
if diff == 1:
return self.step_turn_right()
if diff == 2:
return self.step_turn_right().step_turn_right()
if diff == 3:
return self.step_turn_left()
def go(self, direction):
return self.turn_to(direction).step_go_forward()
def shoot_to(self, direction):
return self.turn_to(direction).step_shoot_arrow()
def step_generator(self):
state = self
if self.cfs is None:
self.cfs = CFS(sensors=state.sensors().keys(),
actions=state.actions().keys())
for k in xrange(2000):
self.cfs.add_message(state.sensors_msg())
action = self.cfs.get_action()
try:
state = state.act(action)
except garlicsim.misc.WorldEnded:
self.cfs.clear_messages()
self.cfs.give_feedback(state.points)
self.cfs.evolve()
state = State()
yield state
raise garlicsim.misc.WorldEnded
@staticmethod
def create_root():
return State()
