def self_play(agent):
s = State()
states = []
actions = []
while s.winner == 0:
batch, valid, _ = state_tensor([s])
idx = agent.act(batch, valid)
a = Action(idx // N, idx % N)
states.append(s.copy())
actions.append(a)
s.make_move(a)
for i in range(len(states)):
states[i].winner = s.winner
return states, actions
def duel(*players, show=False):
s = State()
player = 0
while s.winner == 0:
batch, valid, _ = state_tensor([s])
while True:
agent = players[player]
idx = agent(batch, valid)
a = Action(idx // N, idx % N)
if s.legal_move(a):
break
if show:
s.show_board()
s.make_move(a)
player = (player + 1) % len(players)
return s.winner
