def search(self, state, color, depth, subtree):
"""
path: if you choose 19 from enable=[13,19,20,21] (choose enable[1])
and then opponent choose 32 from enable=[14,24,32,53], (enable[2])
the path is [1,2]
"""
enable = reversi.getPossiblePoints(state, color)
if depth == self.depth + 1:
# no thinking (simulate)
if len(enable) == 0:
return self.simulate(state, 65, color)
row, line = random.choice(enable)
return self.simulate(state, row * 8 + line, color)
if len(enable) == 0:
return self.search(state, color ^ 1, depth + 1, subtree)
if len(subtree) == 0:
# first visit
subtree.extend([[0, 0, []] for _ in enable])
wins = np.array([node[0] for node in subtree])
loses = np.array([node[1] for node in subtree])
values = beta.rvs(wins + 1, loses + 1)
choice = values.argmax()
row, line = enable[choice]
reversi.putStone(state, row, line, color)
r = self.search(state, color ^ 1, depth + 1, subtree[choice][2])
if r == color:
subtree[choice][0] += 1
else:
subtree[choice][1] += 1
return r
def play(self, state):
enable = reversi.getPossiblePoints(state, self.color)
if len(enable) == 0:
return 65
else:
put = random.choice(enable)
return put[0] * 8 + put[1]
def act(self, state):
leftStones = len(np.where(state == -1)[0])
if leftStones <= self.allSearchDepth:
enables = reversi.getPossiblePoints(state, self.color)
if len(enables) == 0:
return 65
myWins = allSearch(state, self.color, enables, 0,
self.allSearchDepth)
choice = np.argmax(myWins)
row, line = enables[choice]
return row * 8 + line
else:
return self.play(state)
def allSearch(state, color, enables, nowDepth, maxDepth):
"""
全探索
enables: list
colorから見た勝ち石の数listを返す
"""
if nowDepth >= maxDepth:
# stop searching
return [
len(np.where(state == color)[0]) -
len(np.where(state == color ^ 1)[0])
]
myWins = []
if len(enables) == 0:
# pass
opponentEnables = reversi.getPossiblePoints(state, color ^ 1)
if len(opponentEnables) == 0:
# game set
myWins.append(
len(np.where(state == color)[0]) -
len(np.where(state == color ^ 1)[0]))
else:
opponentWins = allSearch(state, color ^ 1, opponentEnables,
nowDepth + 1, maxDepth)
mybest = min(opponentWins)
myWins.append(-mybest)
else:
board = deepcopy(state)
for row, line in enables:
reversi.putStone(board, row, line, color)
opponentEnables = reversi.getPossiblePoints(board, color ^ 1)
opponentWins = allSearch(board, color ^ 1, opponentEnables,
nowDepth + 1, maxDepth)
mybest = min(opponentWins)
myWins.append(-mybest)
return myWins
def play(self, state):
enable = reversi.getPossiblePoints(state, self.color)
if len(enable) == 0:
return 65
if len(enable) == 1:
return enable[0][0] * 8 + enable[0][1]
self.tree = []
for i in range(self.maxSize):
tmpState = deepcopy(state)
self.search(tmpState, self.color, 1, self.tree)
wins = np.array([node[0] for node in self.tree])
loses = np.array([node[1] for node in self.tree])
# best = (wins + loses).argmax()
best = beta.median(wins + 1, loses + 1).argmax()
return enable[best][0] * 8 + enable[best][1]
def play(self, state):
enable = reversi.getPossiblePoints(state, self.color)
if len(enable) == 0:
return 65
maxWin = -1
putPoint = None
for row, line in enable:
win = 0
for i in range(self.size):
tmpState = deepcopy(state)
r = self.simulate(tmpState, row * 8 + line, self.color)
if r == self.color: # winner is me
win += 1
if maxWin == -1 or maxWin < win:
maxWin = win
putPoint = row * 8 + line
return putPoint
def act(self, state):
print("\033[K\n" * 2 + "\033[2A", end="")
if len(reversi.getPossiblePoints(state, self.color)) == 0:
# pass turn
print("\rplayer: pass\033[K", end="")
return 65
while True:
print("\r\033[K>>> ", end="")
query = input()
try:
row, line = map(int, query.split(","))
except:
print("input row, line (0-7)\033[1A", end="")
continue
if 0 <= row and row < 8 and 0 <= line and line < 8:
# valid query
print("\033[1A", end="")
return row * 8 + line
else:
print("input row, line (0-7)\033[1A", end="")