def Expand(self, node, game_state):
"""
Function: Adds a random child to node,
Updates game_state and node's untriedMoves
Returns : The created child
"""
untriedMoves = node.untriedMoves
m = choice(untriedMoves)
# node's untriedMoves updated here
node.untriedMoves.remove(m)
# game_state updated here
game_state = HexBoard.makeMove(m, game_state)
# Check the child, if it is already appended return it otherwise create that child
child = node.getChild(game_state)
if child is not None:
return child
child_type = 'MIN'
if node.type == 'MIN':
child_type = 'MAX'
child = Node(node_type=child_type,
board_state=game_state.board,
parent=node)
# node's children updated here
node.children.append(child)
return child
def MCTS_Player(hyperparameters, isMaximizer):
# First Computer's turn (MCTS)
mcts_agent = MCTS(
game=game, cp=hyperparameters[0],
n=hyperparameters[1]) # initialize mcts agent
move, params['node'] = mcts_agent.search(
params['node'], hyperparameters[2], hyperparameters[3],
isMaximizer)
params['game'] = HexBoard.makeMove(move, params['game'])
params['node'] = util.updateNode(params['node'],
params['game'])
def PlayervsComputer(self):
game = self.game
self.clearOutput()
game.print()
node = Node(node_type='MIN', board_state=game.board,
parent=None) # 1 - initialize node
mcts_agent = MCTS(game=game, cp=0.8,
n=100) # 3 - initialize mcts agent
util = UTIL(infinity=np.inf,
maximizer=game.maximizer,
minimizer=game.minimizer) # 4 - initialize util class
while not game.isGameOver():
if (game.turn % 2) == 0:
move = self.getReady(game)
game = HexBoard.makeMove(move, game)
node = util.updateNode(node, game)
else:
print("Computer is thinking!!!")
itermax = 100 # maximum iteration for search
move, node = mcts_agent.search(node,
itermax,
delta=10,
isMaximizer=True)
print(f'best move: {move}')
game = HexBoard.makeMove(move, game)
node = util.updateNode(node, game)
self.clearOutput()
game.print()
if game.isGameOver():
if game.checkWin(game.BLUE):
print("!!! Blue Player Won !!!")
elif game.checkWin(game.RED):
print("!!! Red Player Won !!!")
def IDTT_Player(hyperparameters, isMaximizer):
# iterative deepening with 4 depth-Dijkstra
boardState = params['node'].board.copy()
try:
move = IterativeDeepeningTranspositionTable[
boardState.tobytes()]
except KeyError:
best_value = util.iterativeDeepening(
params['node'], isMaximizer, hyperparameters[0],
hyperparameters[1])
move = util.getBestMove(params['node'], best_value)
IterativeDeepeningTranspositionTable[
boardState.tobytes()] = move
params['game'] = HexBoard.makeMove(move, params['game'])
params['node'] = util.updateNode(params['node'],
params['game'])
def search(self, root, itermax, delta, isMaximizer):
"""Return the best moves based on MCTS"""
end_time = datetime.now() + timedelta(seconds=delta)
# now = time.time()
while datetime.now() < end_time and itermax > 0:
# We don't want to change the game, we will turn to it in each iteration
game_state = deepcopy(self.game)
# we will change the root (will be Expanded)
node = root
path = [root.board.tobytes()]
# Select
# node is fully expanded and non-terminal
while (node.untriedMoves == []) and (node.children != []) and (
not game_state.isTerminal()):
node = self.UCTSelectChild(node, isMaximizer)
m = HexBoard.getMove(game_state.board, node.board)
game_state = HexBoard.makeMove(m, game_state)
path.append(node.board.tobytes())
# Expand
if (node.untriedMoves != []) and (not game_state.isTerminal()):
# node is expanded and updated with child
node = self.Expand(node, game_state)
path.append(node.board.tobytes())
# Playouts
for p in range(self.N):
# for each playout we want to return to same game_state
_game = deepcopy(game_state)
while not _game.isTerminal():
move = choice(HexBoard.getFreeMoves(_game.board))
_game = HexBoard.makeMove(move, _game)
# This works just once for a particular node
if node.wins == np.inf:
node.wins = 0
if node.loss == np.inf:
node.loss = 0
if _game.checkWin(_game.maximizer):
node.wins += 1
else:
node.loss += 1
# print(f'Playout {p} is done!')
# Backpropagate
# We are removing current node from path
path.pop()
while node is not None:
# backpropagate works from the current node to the root node
if len(path) > 0:
parent = node.getParent(path.pop())
if parent.wins == np.inf:
parent.wins = 0
parent.loss = 0
parent.wins += node.wins
parent.loss += node.loss
else:
parent = None
node.visit += 1
node = parent
itermax -= 1
# print(f"Iteration completed!: It took {time.time() - now}s")
sortedList = sorted(root.children, key=lambda c: c.visit)
# return the move that was most visited
return HexBoard.getMove(root.board, sortedList[-1].board), root