def play(moves, move_time):
# add players
gm = GameMaster(get_gdl_for_game("breakthroughSmall"))
for role in gm.sm.get_roles():
gm.add_player(PUCTPlayer(conf=compete), role)
sm = lookup.by_name("breakthroughSmall").get_sm()
sm.reset()
pretty_board(sm)
# get some states
joint_move = sm.get_joint_move()
base_state = sm.get_initial_state()
def to_cords(s):
mapping_x_cord = {x0: x1 for x0, x1 in zip('abcdef', '654321')}
return mapping_x_cord[s[0]], s[1]
def f(ri, i):
return sm.legal_to_move(ri, ls.get_legal(i))
lead_role_index = 0
gdl_moves = []
for m in moves:
from_, to_ = map(to_cords, m.split("-"))
gdl_move = []
for ri in range(2):
if ri == lead_role_index:
gdl_role_move = "(move %s %s %s %s)" % (from_[0], from_[1],
to_[0], to_[1])
else:
gdl_role_move = "noop"
ls = sm.get_legal_state(ri)
the_moves = [f(ri, ii) for ii in range(ls.get_count())]
choice = the_moves.index(gdl_role_move)
joint_move.set(ri, ls.get_legal(choice))
gdl_move.append(str(gdl_role_move))
# update state machine
sm.next_state(joint_move, base_state)
sm.update_bases(base_state)
lead_role_index ^= 1
gdl_moves.append(str(gdl_move))
print "%s -> %s" % (m, gdl_move)
pretty_board(sm)
# play move via gamemaster:
gm.reset()
gm.start(meta_time=15,
move_time=move_time,
initial_basestate=sm.get_current_state(),
game_depth=len(moves))
move = gm.play_single_move(last_move=None)
player = gm.get_player(lead_role_index)
move = move[lead_role_index]
move = move.replace("(move", "").replace(")", "")
a, b, c, d = move.split()
mapping_x_cord = {x0: x1 for x0, x1 in zip('654321', 'abcdef')}
next_move = "%s%s-%s%s" % (mapping_x_cord[a], b, mapping_x_cord[c], d)
sm.update_bases(gm.sm.get_current_state())
pretty_board(sm)
print "PLAYED", next_move, player.last_probability
return next_move, player.last_probability
class GameMasterByGame(object):
def __init__(self, game_config):
assert isinstance(game_config, GameConfig)
self.game_config = game_config
game_name = self.game_config.game_name
if game_name == "breakthrough2":
game_name = "breakthrough"
self.gm = GameMaster(get_gdl_for_game(game_name))
# add players
puct_conf = self.get_puct_config()
for role in self.gm.sm.get_roles():
self.gm.add_player(puctplayer.PUCTPlayer(conf=puct_conf), role)
def get_puct_config(self):
multiplier = self.game_config.sims_multiplier
if multiplier == 0:
playouts_per_iteration = 1
else:
playouts_per_iteration = 100 * multiplier
conf = confs.PUCTPlayerConfig(
name="clx",
generation=self.game_config.generation,
verbose=True,
playouts_per_iteration=playouts_per_iteration,
playouts_per_iteration_noop=0,
dirichlet_noise_alpha=-1,
root_expansions_preset_visits=-1,
fpu_prior_discount=0.25,
puct_before_expansions=3,
puct_before_root_expansions=4,
puct_constant_before=3.0,
puct_constant_after=0.75,
choose="choose_temperature",
temperature=1.5,
depth_temperature_max=self.game_config.depth_temperature_max,
depth_temperature_start=4,
depth_temperature_increment=0.25,
depth_temperature_stop=self.game_config.depth_temperature_stop,
random_scale=0.9,
max_dump_depth=2)
return conf
def wtf(self, str_state):
bs = self.gm.convert_to_base_state(str_state)
nn = self.gm.get_player(0).nn
print nn.predict_1(bs.to_list())
def get_move(self, str_state, depth, lead_role_index):
print "GameMasterByGame.get_move", str_state
self.gm.reset()
self.gm.start(
meta_time=120,
move_time=120,
initial_basestate=self.gm.convert_to_base_state(str_state),
game_depth=depth)
move = self.gm.play_single_move(last_move=None)
player = self.gm.get_player(lead_role_index)
return move[
lead_role_index], player.last_probability, self.gm.finished()