def traverse(self, node: TreeNode, board: Board):
"""
Expand node.
:param node: Current node.
:param board: The board.
:return: <TreeNode> Expanded nodes.
"""
while True:
if len(node.children) == 0:
break
action, node = node.choose_best_child(c=self.greedy_value)
board.step(action)
is_over, _ = board.result()
if is_over:
return node
# Expand all child node.
actions = board.available_actions
probs = np.ones(len(actions)) / len(actions)
for action, prob in zip(actions, probs):
_ = node.expand(action, prob)
return node
def traverse(self, node: TreeNode, board: Board):
"""
扩展子节点。
Expand node.
:param node: 当前节点。 Current node.
:param board: 棋盘。 The board.
:return: (<TreeNode>, value<int>) 扩展出的节点和需要反向传输的 value。
Expanded nodes, and the value to be backpropagated.
"""
while True:
if len(node.children) == 0:
break
action, node = node.choose_best_child(c=self.greedy_value)
board.step(action)
# 是否结束。 game over?
is_over, winner = board.result()
if is_over:
if winner == board.current_player:
value = 1.0
elif winner == -board.current_player:
value = -1.0
else:
value = 0.0
return node, value
# 使用策略价值函数决策当前动作概率及评估价值。
# Use the strategy value function to decide the current action probability and evaluate the value.
action_probs, value = self.policy_value_function(board)
for action, probability in action_probs:
_ = node.expand(action, probability)
return node, value
def traverse(self, node: TreeNode, board: Board):
"""
扩展子节点。
Expand node.
:param node: 当前节点。 Current node.
:param board:
:return: <TreeNode> 扩展出的节点。 Expanded nodes.
"""
while True:
is_over, _ = board.result()
if is_over:
break
if len(node.children) != 0:
action, node = node.choose_best_child(c=5.0)
board.step(action)
else:
actions = board.available_actions
probs = np.ones(len(actions)) / len(actions)
# 扩展所有子节点。 Expand all child node.
for action, prob in zip(actions, probs):
_ = node.expand(action, prob)
break
return node, board