def train(debug, iterations, table):
game = TicTacToe()
ai1 = Agent(True, table)
ai2 = Agent(True, table)
try:
while True:
if iterations == 0:
table.save_q_table()
break
if iterations > 0:
#if iterations < 50:
# debug = True
iterations -= 1
if iterations % 100_000 == 0:
print(iterations)
print("q_table len", len(table.table))
if game.is_board_full() or game.get_winner():
if game.get_winner() is Player.ONE:
ai1.reward(1)
ai2.reward(0)
elif game.get_winner() is Player.TWO:
ai1.reward(0)
ai2.reward(1)
else:
ai1.reward(0.1)
ai2.reward(0.5)
game.reset()
ai1.reset_history()
ai2.reset_history()
continue
ai1.iterate(game, print_q=debug) if game.get_player(
) == Player.ONE else ai2.iterate(game)
if debug:
print()
print(game.get_hash())
game.print_board()
time.sleep(0.1)
continue
except KeyboardInterrupt:
table.save_q_table()
exit()
def iterate(self,
game: TicTacToe,
train: bool = False,
print_q: bool = False):
moves = game.get_legal_moves()
move = self.random_move(moves)
field_hash = game.get_hash()
if self.ai:
move = self.get_optimal_move(moves, field_hash, print_q, train)
self.history = [{
"hash": field_hash,
"move": str(move)
}] + self.history
game.input(int(move))