训练麻将打时候报错,看起来像是在已经

[acracker]

看起来像是dealer.deck中已经没有牌了， 但是还会执行step， 我看了一下代码，也不知道如何退出。

err:

C:\Users\pl\MiniConda3\python.exe D:/code/mahjong/t2.py

Logs saved in experiments/leduc_holdem_cfr_result/
Traceback (most recent call last):
  File "D:/code/mahjong/t2.py", line 131, in <module>
    train(args)
  File "D:/code/mahjong/t2.py", line 65, in train
    agent.train()
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\agents\cfr_agent.py", line 41, in train
    self.traverse_tree(probs, player_id)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\agents\cfr_agent.py", line 73, in traverse_tree
    utility = self.traverse_tree(new_probs, player_id)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\agents\cfr_agent.py", line 73, in traverse_tree
    utility = self.traverse_tree(new_probs, player_id)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\agents\cfr_agent.py", line 73, in traverse_tree
    utility = self.traverse_tree(new_probs, player_id)
  [Previous line repeated 86 more times]
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\agents\cfr_agent.py", line 72, in traverse_tree
    self.env.step(action)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\envs\env.py", line 84, in step
    next_state, player_id = self.game.step(action)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\games\mahjong\game.py", line 68, in step
    self.round.proceed_round(self.players, action)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\games\mahjong\round.py", line 78, in proceed_round
    self.dealer.deal_cards(players[self.current_player], 1)
  File "C:\Users\pl\MiniConda3\lib\site-packages\rlcard\games\mahjong\dealer.py", line 26, in deal_cards
    player.hand.append(self.deck.pop())
IndexError: pop from empty list

code:

# -*- coding: utf-8 -*-
import os
import logging
import argparse

import rlcard
from rlcard.agents import (
    CFRAgent,
    RandomAgent,
)
from rlcard.utils import (
    set_seed,
    tournament,
    Logger,
    plot_curve,
)


def train(args):
    # Make environments, CFR only supports Leduc Holdem
    env = rlcard.make(
        args.env,
        config={
            'seed': 0,
            'allow_step_back': True,
        }
    )
    eval_env = rlcard.make(
        'leduc-holdem',
        config={
            'seed': 0,
        }
    )

    # Seed numpy, torch, random
    set_seed(args.seed)

    # Initilize CFR Agent
    agent = CFRAgent(
        env,
        os.path.join(
            args.log_dir,
            'cfr_model',
            args.env,
        ),
    )
    agent.load()  # If we have saved model, we first load the model

    # Evaluate CFR against random
    eval_env.set_agents([
        agent,
        RandomAgent(num_actions=env.num_actions),
    ])

    # Start training
    with Logger(args.log_dir) as logger:
        for episode in range(args.num_episodes):
            agent.train()
            print('\rIteration {}'.format(episode), end='')
            # Evaluate the performance. Play with Random agents.
            if episode % args.evaluate_every == 0:
                agent.save() # Save model
                logger.log_performance(
                    episode,
                    tournament(
                        eval_env,
                        args.num_eval_games
                    )[0]
                )

        # Get the paths
        csv_path, fig_path = logger.csv_path, logger.fig_path
    # Plot the learning curve
    plot_curve(csv_path, fig_path, 'cfr')


if __name__ == '__main__':
    parser = argparse.ArgumentParser("CFR example in RLCard")
    parser.add_argument(
        '--env',
        type=str,
        default='mahjong',
        choices=[
            'blackjack',
            'leduc-holdem',
            'limit-holdem',
            'doudizhu',
            'mahjong',
            'no-limit-holdem',
            'uno',
            'gin-rummy',
            'bridge',
        ],
    )

    parser.add_argument(
        '--seed',
        type=int,
        default=42,
    )
    parser.add_argument(
        '--num_episodes',
        type=int,
        default=5000,
    )
    parser.add_argument(
        '--num_eval_games',
        type=int,
        default=2000,
    )
    parser.add_argument(
        '--evaluate_every',
        type=int,
        default=100,
    )
    parser.add_argument(
        '--log_dir',
        type=str,
        default='experiments/leduc_holdem_cfr_result/',
    )

    args = parser.parse_args()
    logging.basicConfig(level=logging.DEBUG)
    train(args)