def resume_self_play():
env = TicTacToeEnv()
saves = [f for f in listdir(save_dir) if isfile(join(save_dir, f))]
recent_file = max(saves)
policy = EpsilonGreedy(QConvTicTacToe(env), 0)
opposing_policy = EpsilonGreedy(QConvTicTacToe(env), 1)
self_play = SelfPlay(policy, opposing_policy)
policy.q.policy_net.load_state_dict(torch.load(join(save_dir, recent_file)))
self_play.evaluate_policy(100)
def resume_self_play():
env = Connect4Env()
saves = [f for f in listdir(save_dir) if isfile(join(save_dir, f))]
recent_file = max(saves)
policy = EpsilonGreedy(QLinear(env), 0)
opposing_policy = EpsilonGreedy(QLinear(env), 0) # Acts greedily
self_play = SelfPlay(policy, opposing_policy)
policy.q.policy_net.load_state_dict(torch.load(join(save_dir, recent_file)))
self_play.evaluate_policy(100)