def run(display, retrain, num_episodes):
pygame.init()
env = SnakeEnv()
agent = QlearningAgent(env)
if not retrain:
try:
load_policy(agent)
except:
pass
for _ in tqdm(range(num_episodes)):
state = env.reset()
done = False
while not done:
for event in pygame.event.get():
if event.type == pygame.QUIT:
save_policy(agent)
pygame.quit()
sys.exit()
if display:
env.render()
action = agent.act(state)
next_state, reward, done = env.step(action)
agent.update_q_value(state, reward, action, next_state, done)
state = next_state
agent.epsilon = max(agent.epsilon * agent.epsilon_decay_rate,
agent.min_epsilon)
save_policy(agent)
def show_gen(idx):
global ginst
genaction = GenAction(idx)
env = SnakeEnv(use_simple=True)
obs = env.reset()
c = 0
while True:
c += 1
action = genaction(obs)
_, obs, done, _ = env(action)
extra = '代数: {}'.format(ginst.n_gen)
if c % 3 == 0:
if ginst.n_gen:
env.render(extra)
else:
env.render()
# sleep(1 / ginst.n_gen)
print(done, env.status.direction, env.life)
if done:
break
sleep(1.5)
def show_gen(idx):
global ginst
genaction = GenAction(idx)
env = SnakeEnv(use_simple=True)
obs = env.reset()
while True:
action = genaction(obs)
_, obs, done, _ = env(action)
env.render()
if done:
break
env.close()
from stable_baselines.common.vec_env import DummyVecEnv
from stable_baselines import PPO2
from snake_env import SnakeEnv
parser = argparse.ArgumentParser()
parser.add_argument('--mode', choices=['train', 'test'], default='test')
args = parser.parse_args()
env = SnakeEnv((20, 20), 'standard')
env = DummyVecEnv([lambda: env])
model = PPO2(CnnPolicy, env, verbose=1)
if args.mode == 'train':
model.learn(total_timesteps=20000)
model.save('policy_baseline_snake')
elif args.mode == 'test':
obs = env.reset()
model.load('policy_baseline_snake')
for i in range(1000):
action, _states = model.predict(obs)
obs, reward, done, info = env.step(action)
env.render()
if done:
env.reset()
env.close()
def play(record=0, no_render=False):
env = SnakeEnv(need_render=not no_render, alg='最短路径')
obs = env.reset()
env.render()
input()
x, y = [], []
directions = {
(-1, 0): env.right,
(1, 0): env.left,
(0, -1): env.down,
(0, 1): env.up
}
need_record = True if record else False
new_dst = None
origin_dst = None
# counter = 20
use_random = False
while True:
if not record and not no_render:
env.render()
src = np.where(obs == 2)
src = int(src[1]), int(src[0])
dst = np.where(obs == -1)
dst = int(dst[0]), int(dst[1])
if new_dst is not None:
paths = bfs(obs, start=src, dst=new_dst)
else:
paths = bfs(obs, start=src, dst=dst)
if paths is None:
# origin_dst = dst
# new_dst = (
# np.random.randint(0, obs.shape[0]),
# np.random.randint(0, obs.shape[1]),
# )
# counter -= 1
# if counter <= 0:
# print('score: ', env.status.score)
# new_dst = None
# origin_dst = None
# counter = 20
# obs = env.reset()
# continue
use_random = True
else:
new_dst = None
if new_dst is not None and paths[1] == new_dst:
new_dst = None
if origin_dst is not None:
dst = origin_dst
origin_dst = None
# counter = 20
continue
# if counter <= 0 or paths is None or len(paths) <= 1:
# print('score: ', env.status.score)
# obs = env.reset()
# continue
if use_random:
action = np.random.randint(0, 4)
use_random = False
else:
dst = paths[1]
dire = src[0] - dst[0], src[1] - dst[1]
action = directions[dire]
# import ipdb
# ipdb.set_trace()
if need_record:
x.append(obs)
y.append(action)
if len(y) >= record:
return x, y
if len(y) % 1000 == 0:
print(len(y))
_, obs, done, _ = env(action)
# counter = 20
if done:
print(env.status.score)
sleep(1.5)
break
if not record and not no_render:
env.render()
env.close()
