def genmove(self, game_state, player):
self.move_num += 1
dist = self.bot.gen_probdist(game_state, player)
result = gtp_states.Move_generator_result()
if dist is not None:
move = np.unravel_index(np.argmax(dist), dist.shape)
result.move = move
logging.debug("%s valid moves\n%s"%(self,
utils.dist_stats(dist)))
logging.debug("%s move %d: playing %s"%(self,
self.move_num,
gomill.common.format_vertex(move)))
else:
result.pass_move = True
logging.debug("%s move %d: playing pass"%(self, self.move_num))
return result
def genmove(self, game_state, player):
self.move_num += 1
dist = self.bot.gen_probdist(game_state, player)
result = gtp_states.Move_generator_result()
if dist is not None:
move = np.unravel_index(np.argmax(dist), dist.shape)
result.move = move
logging.debug("%s valid moves\n%s" %
(self, utils.dist_stats(dist)))
logging.debug(
"%s move %d: playing %s" %
(self, self.move_num, gomill.common.format_vertex(move)))
else:
result.pass_move = True
logging.debug("%s move %d: playing pass" % (self, self.move_num))
return result
def test_bot():
logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s',
level=logging.DEBUG)
player = DistWrappingMaxPlayer(RandomDistBot())
class State:
pass
s = State()
b = gomill.boards.Board(3)
s.board = b
b.play(1, 1, "b")
b.play(0, 1, "b")
logging.debug("\n"+gomill.ascii_boards.render_board(b))
mv = player.genmove(s, 'w').move
b.play(mv[0], mv[1], 'w')
logging.debug("\n"+gomill.ascii_boards.render_board(b))
logging.debug("best move is " + gomill.common.format_vertex(mv))
logging.debug("\n" + str(player.bot.last_dist))
logging.debug(utils.dist_stats(player.bot.last_dist))
def test_bot():
logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s',
level=logging.DEBUG)
player = DistWrappingMaxPlayer(RandomDistBot())
class State:
pass
s = State()
b = gomill.boards.Board(3)
s.board = b
b.play(1, 1, "b")
b.play(0, 1, "b")
logging.debug("\n" + gomill.ascii_boards.render_board(b))
mv = player.genmove(s, 'w').move
b.play(mv[0], mv[1], 'w')
logging.debug("\n" + gomill.ascii_boards.render_board(b))
logging.debug("best move is " + gomill.common.format_vertex(mv))
logging.debug("\n" + str(player.bot.last_dist))
logging.debug(utils.dist_stats(player.bot.last_dist))
def gen_probdist(self, game_state, player):
"""
Generates a correct move probability distribution for the next move,
using the gen_probdist_raw().
Correct means that it zeroes out probability of playing incorrect move,
such as move forbidden by ko, suicide and occupied points.
Stores the dist and the player.
:return: a numpy array of floats of shape (board.side, board.side), or None for pass
the array is normalized to 1
"""
dist = self.gen_probdist_raw(game_state, player)
if dist is not None:
correct_moves = analyze_board.board2correct_move_mask(
game_state.board, player)
if game_state.ko_point:
correct_moves[game_state.ko_point[0]][
game_state.ko_point[1]] = 0
# compute some debugging stats of the incorrect moves first
incorrect_dist = (1 - correct_moves) * dist
logging.debug("%s incorrect moves\n%s" %
(self, utils.dist_stats(incorrect_dist)))
# keep only correct moves
dist = correct_moves * dist
s = dist.sum()
if s > 0.0:
dist = dist / dist.sum()
else:
logging.debug("No valid moves, PASSING.")
dist = None
self.last_dist = dist
self.last_player = player
return self.last_dist
def gen_probdist(self, game_state, player):
"""
Generates a correct move probability distribution for the next move,
using the gen_probdist_raw().
Correct means that it zeroes out probability of playing incorrect move,
such as move forbidden by ko, suicide and occupied points.
Stores the dist and the player.
:return: a numpy array of floats of shape (board.side, board.side), or None for pass
the array is normalized to 1
"""
dist = self.gen_probdist_raw(game_state, player)
if dist is not None:
correct_moves = analyze_board.board2correct_move_mask(game_state.board, player)
if game_state.ko_point:
correct_moves[game_state.ko_point[0]][game_state.ko_point[1]] = 0
# compute some debugging stats of the incorrect moves first
incorrect_dist = (1 - correct_moves) * dist
logging.debug("%s incorrect moves\n%s"%(self,
utils.dist_stats(incorrect_dist)))
# keep only correct moves
dist = correct_moves * dist
s = dist.sum()
if s > 0.0:
dist = dist / dist.sum()
else:
logging.debug("No valid moves, PASSING.")
dist = None
self.last_dist = dist
self.last_player = player
return self.last_dist
def dist_stats(self, top=3):
if self.last_dist is not None:
return utils.dist_stats(self.last_dist, top)
return ''
def dist_stats(self, top=3):
if self.last_dist is not None:
return utils.dist_stats(self.last_dist, top)
return ''
