def MCTS_Search(board, player, num_reads, n_net):
root = Node(board,
checkers.get_all_moves(board, player),
player,
move=None,
parent=ParentRootNode())
for i in range(num_reads):
leaf = root.select_leaf()
player = checkers.switch_player(player)
child_prior_prob, value = n_net(
checkers.get_state2(leaf.board, leaf.player))
# print(child_prior_prob)
# print("The number of reads", i)
if checkers.isTerminal(board) or checkers.get_all_moves(
leaf.board, leaf.player) == []:
print("Finished Game")
leaf.backpropagate(value)
leaf.print_tree()
else:
child_prior_prob = child_prior_prob.cpu().detach().numpy().reshape(
-1)
leaf.expand_and_evaluate(child_prior_prob)
leaf.backpropagate(value)
root.print_tree()
return root
def MCTS_self_play(nnet, num_games, s_index, iteration):
data_x = []
for itt in tqdm(range(s_index, num_games + s_index)):
board = checkers.initial_board(board_size, board_size)
#board = checkers.initial_b6()
player = 1
data = []
value = 0
num_moves = 0
t = 1
while checkers.isTerminal(board, player) is not True:
# if num_moves > 15:
# t = 0.1
root = MCTS_Search(board, player, roll_out, nnet)
# print("The turn of player {:d} and Moves {:d}".format(player, num_moves))
# checkers.print_board(root.board)
policy = get_policy(root, t)
data.append([board, player, policy])
move = np.argmax(policy)
board = checkers.apply_move(root.board, root.possible_moves[move][0], root.possible_moves[move][1],
root.player)
player = checkers.switch_player(player)
if len(checkers.get_all_moves(board, player)) == 0:
# Player == 1 means White pieces
# print("Game Finished")
if player == 1:
value = -1
elif player == 2:
value = 1
else:
value = 0
break
if num_moves == 150:
value = 0
break
num_moves += 1
for ind, dx in enumerate(data):
s, pl, po = dx
if ind == 0:
data_x.append([checkers.get_state2(s, pl), po, 0])
else:
data_x.append([checkers.get_state2(s, pl), po, value])
del data
# filename = "MCTS_iteration-{:d}_game-{:d}.p".format(iteration, itt)
# save_data(filename, data_x)
return data_x
def MCTS_self_play(nnet, num_games, s_index, iteration):
for itt in tqdm(range(s_index, num_games + s_index)):
board = checkers.initial_board(board_size, board_size)
player = 1
data = []
value = 0
num_moves = 0
t = 1
while checkers.isTerminal(board):
if num_moves > 15:
t = 0.1
root = MCTS_Search(board, player, 500, nnet)
policy = get_policy(root, t)
data.append([board, player, policy])
move = np.argmax(policy)
board = checkers.apply_move(root.board,
root.possible_moves[move][0],
root.possible_moves[move][1],
root.player)
player = checkers.switch_player(player)
if len(checkers.get_all_moves(board, player)) == 0:
# Player == 1 means White pieces
if player == 1:
value = 1
elif player == 2:
value = -1
else:
value = 0
if num_moves == 150:
value = 0
break
num_moves += 1
data_x = []
for ind, d in enumerate(data):
s, pl, po = d
if ind == 0:
data_x.append([s, pl, po, 0])
else:
data_x.append([s, pl, po, value])
del data
filename = "MCTS_iteration-{:d}_game-{:d}.p".format(iteration, itt)
save_data(filename, data_x)
return
def MCTS_Play_WithRandom(nnet, num_games):
number_of_wins = 0
number_of_draws = 0
for itt in tqdm(range(num_games)):
board = checkers.initial_board(board_size, board_size)
# board = checkers.initial_b6()
player = 1
num_moves = 0
t = 1
while checkers.isTerminal(board, player) is not True:
# if num_moves > 15:
# t = 0.1
if player == 1:
root = MCTS_Search(board, player, roll_out, nnet)
policy = get_policy(root, t)
move = np.argmax(policy)
board = checkers.apply_move(root.board, root.possible_moves[move][0], root.possible_moves[move][1],
root.player)
else:
move = checkers.get_random_move(board, player)
board = checkers.apply_move(board, move[0], move[1], player)
# print("The turn of player {:d} and Moves {:d}".format(player, num_moves))
# checkers.print_board(board)
player = checkers.switch_player(player)
if len(checkers.get_all_moves(board, player)) == 0:
# Player == 1 means White pieces
# print("Game Finished")
if player == 2:
number_of_wins += 1
break
if num_moves == 200:
number_of_draws += 1
break
num_moves += 1
return number_of_wins, number_of_draws