0

I use Stable_baseline3.PPO to train an agent on highway-fast-v0 (continues action type), and find that when calling ppo.learn() method, it is aborted with "Process finished with exit code 139" and no other error message. And this error is not occur at the same time_step when training, how can I solve it?

import gym 
from stable_baselines3 import PPO
import warnings
warnings.filterwarnings('ignore')
# ==================================
#        Main script
# ==================================

def make_configure_env(**kwargs):
    env = gym.make(kwargs["id"])
    env.configure(kwargs["config"])
    env.reset()
    return env


env_kwargs = {
    'id': 'highway-fast-v0',
    'config': {
        "action": {
            "type": "ContinuousAction"
        }
    }
}
n_cpu = 6
batch_size = 64
env = make_configure_env(**env_kwargs)
env.reset()
model = PPO("MlpPolicy",
            env,
            policy_kwargs=dict(net_arch=[dict(pi=[256, 256], vf=[256, 256])]),
            n_steps=batch_size * 12 // n_cpu,
            batch_size=batch_size,
            n_epochs=10,
            learning_rate=5e-4,
            gamma=0.8,
            verbose=2,
            tensorboard_log="highway_ppo/")
# Train the agent
model.learn(total_timesteps=2e4)
# Save the agent
model.save("highway_ppo_continues/model")
Plankson
  • 1

1 Answers1

0

On reading the code, I'm seeing import highway_env missing in it. I tried using the same code along with import and it was working for me.

Francis Sunny
  • 36
  • 1
  • 6