MC_EXPLORATION_CONSTANT,
a_net=ANET,
epsilon=EPSILON)
for i in tqdm(range(EPISODES + 1)):
# No action needed to reach initial state
action = None
state = mc_game.get_simple_state()
# Init Monte Carlo root
root = Node(state, player, None, action,
mc_game.get_reversed_binary())
while not actual_game.is_terminal_state():
if i in DISPLAY_INDICES:
visualizer.draw(actual_game.get_state(), DISPLAY_DELAY)
# Find the best move using MCTS
new_root, prev_root_children = mc.tree_search(
root, MC_NUMBER_SEARCH_GAMES)
# Distribution of visit counts along all arcs emanating from root
D = [
child.visits / root.visits for child in prev_root_children
]
# Add case to RBUF
RBUF.add(root.state, root.reversed_state, D)
def tournament(self):
self.init_AI()
for players in self.vs:
print(
f"Model {self.model_postfixes[players[0]]} vs. model {self.model_postfixes[players[1]]}"
)
starter = 1
game = HexGame(self.size, starter)
visualizer = HexMapVisualizer(game.board.cells.values(),
True,
self.size,
game_type="hex")
smarter_wins = 0
for i in range(self.number_of_games):
player = self.AI[players[i % 2]]
strtr = player.name
#print(f"Player {game.playing} (model {player.name}) is starting!")
while not game.is_terminal_state():
#print(f"Player {player.name} moving")
if self.display:
visualizer.draw(game.get_state(), self.delay)
state = game.get_simple_state()
legal_moves = game.get_reversed_binary()
possible_states = game.generate_possible_child_states()
pred, idx = player.model.get_move(state, legal_moves)
if random() > (0 if self.display else 0.2):
best_index = idx
else:
best_index = np.random.choice(np.arange(len(pred)),
p=pred)
data = possible_states[best_index]
game.do_action(data["action"])
prev_player = player.name
player = self.change_turn(players, player)
if self.display:
visualizer.draw(game.get_state(), self.delay * 3)
#print(f"Model {strtr} started, and model {prev_player} (player {game.playing}) won!")
smarter = self.model_postfixes[players[1]]
if prev_player == smarter:
smarter_wins += 1
#starter = 2 if starter == 1 else 1
game.reset(starter, hard=True)
print(
f"Model {smarter} won {smarter_wins} out of {self.number_of_games} games ({smarter_wins/self.number_of_games*100}%)"
)