Use Closed-Form Policy to Play Blackjack-v1¶

In [1]:
import sys
import logging
import itertools

import numpy as np
np.random.seed(0)
import gym

logging.basicConfig(level=logging.INFO,
        format='%(asctime)s [%(levelname)s] %(message)s',
        stream=sys.stdout, datefmt='%H:%M:%S')
In [2]:
env = gym.make('Blackjack-v1')
for key in vars(env):
    logging.info('%s: %s', key, vars(env)[key])
for key in vars(env.spec):
    logging.info('%s: %s', key, vars(env.spec)[key])

00:00:00 [INFO] env: <BlackjackEnv<Blackjack-v1>>
00:00:00 [INFO] _action_space: None
00:00:00 [INFO] _observation_space: None
00:00:00 [INFO] _reward_range: None
00:00:00 [INFO] _metadata: None
00:00:00 [INFO] _has_reset: False
00:00:00 [INFO] entry_point: gym.envs.toy_text:BlackjackEnv
00:00:00 [INFO] reward_threshold: None
00:00:00 [INFO] nondeterministic: False
00:00:00 [INFO] max_episode_steps: None
00:00:00 [INFO] order_enforce: True
00:00:00 [INFO] kwargs: {'sab': True, 'natural': False}
00:00:00 [INFO] namespace: None
00:00:00 [INFO] name: Blackjack
00:00:00 [INFO] version: 1
In [3]:
class ClosedFormAgent:
    def __init__(self, _):
        pass

    def reset(self, mode=None):
        pass

    def step(self, observation, reward, terminated):
        player, dealer, has_ace = observation
        if player <= 11:  # always draw, never bust
            action = 1
        elif has_ace:
            action = (player <= 17) or ((player == 18) and (dealer in [1, 2, 10]))
        elif player >= 17:
            action = 0
        else:
            action = [
                [1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1],
                [1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1],
                [1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1],
                [1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1],
                [1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1],
            ][player - 12][dealer]
        return action

    def close(self):
        pass


agent = ClosedFormAgent(env)
In [4]:
def play_episode(env, agent, seed=None, mode=None, render=False):
    observation, _ = env.reset(seed=seed)
    reward, terminated, truncated = 0., False, False
    agent.reset(mode=mode)
    episode_reward, elapsed_steps = 0., 0
    while True:
        action = agent.step(observation, reward, terminated)
        if render:
            env.render()
        if terminated or truncated:
            break
        observation, reward, terminated, truncated, _ = env.step(action)
        episode_reward += reward
        elapsed_steps += 1
    agent.close()
    return episode_reward, elapsed_steps


logging.info('==== test ====')
episode_rewards = []
for episode in range(100):
    episode_reward, elapsed_steps = play_episode(env, agent)
    episode_rewards.append(episode_reward)
    logging.info('test episode %d: reward = %.2f, steps = %d',
            episode, episode_reward, elapsed_steps)
logging.info('average episode reward = %.2f ± %.2f',
        np.mean(episode_rewards), np.std(episode_rewards))

00:00:00 [INFO] ==== test ====
00:00:00 [INFO] test episode 0: reward = -1.00, steps = 3
00:00:00 [INFO] test episode 1: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 2: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 3: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 4: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 5: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 6: reward = -1.00, steps = 3
00:00:00 [INFO] test episode 7: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 8: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 9: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 10: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 11: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 12: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 13: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 14: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 15: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 16: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 17: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 18: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 19: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 20: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 21: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 22: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 23: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 24: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 25: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 26: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 27: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 28: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 29: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 30: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 31: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 32: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 33: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 34: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 35: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 36: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 37: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 38: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 39: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 40: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 41: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 42: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 43: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 44: reward = -1.00, steps = 3
00:00:00 [INFO] test episode 45: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 46: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 47: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 48: reward = 1.00, steps = 3
00:00:00 [INFO] test episode 49: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 50: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 51: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 52: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 53: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 54: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 55: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 56: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 57: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 58: reward = -1.00, steps = 3
00:00:00 [INFO] test episode 59: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 60: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 61: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 62: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 63: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 64: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 65: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 66: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 67: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 68: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 69: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 70: reward = 1.00, steps = 3
00:00:00 [INFO] test episode 71: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 72: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 73: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 74: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 75: reward = -1.00, steps = 2
00:00:00 [INFO] test episode 76: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 77: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 78: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 79: reward = 0.00, steps = 2
00:00:00 [INFO] test episode 80: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 81: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 82: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 83: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 84: reward = -1.00, steps = 4
00:00:00 [INFO] test episode 85: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 86: reward = 1.00, steps = 3
00:00:00 [INFO] test episode 87: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 88: reward = -1.00, steps = 3
00:00:00 [INFO] test episode 89: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 90: reward = 0.00, steps = 1
00:00:00 [INFO] test episode 91: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 92: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 93: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 94: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 95: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 96: reward = 1.00, steps = 2
00:00:00 [INFO] test episode 97: reward = -1.00, steps = 1
00:00:00 [INFO] test episode 98: reward = 1.00, steps = 1
00:00:00 [INFO] test episode 99: reward = 1.00, steps = 1
00:00:00 [INFO] average episode reward = -0.14 ± 0.95
In [5]:
env.close()