def dp(self, player, labels):
id = self.state_to_id(labels)
if not id in self.sg:
candidates = model_api.get_candidates(player, labels, self.graph_mat)
sg_set = set()
if candidates:
for c in candidates:
labels[c] = player
sg_set.add(self.dp(3-player, labels))
labels[c] = 0
self.sg[id] = self.mex(sg_set)
return self.sg[id]
def forward(self, obs):
player, labels = obs
candidates = model_api.get_candidates(player, labels, self.graph_mat)
#print(player, labels, candidates)
if candidates:
for c in candidates:
labels[c] = player
if self.sg[self.state_to_id(labels)] == 0:
labels[c] = 0
return c
labels[c] = 0
return random.sample(candidates, 1)[0]
else:
return random.randint(0, self.n-1)
def forward(self, obs):
player, labels = obs
candidates = model_api.get_candidates(player, labels, self.graph_mat)
if candidates:
for c in candidates:
labels[c] = player
values = [
self.get_value(player, labels, self.graph_mat,
self.graph_dis) for c in candidates
]
labels[c] = 0
return candidates[np.argmax(values)]
else:
return random.randint(0, self.n - 1)
def min_max_search(self, depth, player, labels):
candidates = model_api.get_candidates(player, labels, self.graph_mat)
if not candidates:
return -config.inf, None
max_score = -config.inf
policy = None
for c in candidates:
labels[c] = player
if depth < self.max_depth:
score, _ = self.min_max_search(depth+1, 3-player, labels)
score = -score
else:
score = self.get_value(player, labels, self.graph_mat, self.graph_dis)
labels[c] = 0
if score > max_score:
max_score = score
policy = c
return max_score, policy
def ab_search(self, depth, player, labels, history_max):
candidates = model_api.get_candidates(player, labels, self.graph_mat)
if not candidates:
return -config.inf, None
max_score = -config.inf
policy = None
random.shuffle(candidates)
for c in candidates:
labels[c] = player
if depth < self.max_depth:
score, _ = self.ab_search(depth + 1, 3 - player, labels,
min(history_max, -max_score))
score = -score
else:
score = self.get_value(player, labels, self.graph_mat,
self.graph_dis)
labels[c] = 0
if score > max_score:
max_score = score
policy = c
if max_score >= history_max:
return max_score, policy
return max_score, policy