adaa8aefa8
* fixes to enjoy_cartpole, enjoy_mountaincar.py
* fixed {train,enjoy}_pong, removed enjoy_retro
* set number of timesteps to 1e7 in train_pong
* flake8 complaints
* use synchronous version fo acktr in test_env_after_learn
* flake8
29 lines
625 B
Python
29 lines
625 B
Python
import gym
|
|
from baselines import deepq
|
|
|
|
|
|
def main():
|
|
env = gym.make("PongNoFrameskip-v4")
|
|
env = deepq.wrap_atari_dqn(env)
|
|
model = deepq.learn(
|
|
env,
|
|
"conv_only",
|
|
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
|
|
hiddens=[256],
|
|
dueling=True,
|
|
total_timesteps=0
|
|
)
|
|
|
|
while True:
|
|
obs, done = env.reset(), False
|
|
episode_rew = 0
|
|
while not done:
|
|
env.render()
|
|
obs, rew, done, _ = env.step(model(obs[None])[0])
|
|
episode_rew += rew
|
|
print("Episode reward", episode_rew)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|