我正在使用rllib,并希望训练DQN代理。我想每隔N = 10集存储一次在测试环境和培训环境中播放代理的视频。可以在config中做到吗?
tune.run(
"DQN",
stop={
"timesteps_total": 200000,
# "episode_reward_mean": 200
},
config={
"env": 'MyEnv', # ( I have registered my ENV)
"model": {
"custom_model": "my_model",
},
"lr": 1e-4,
"num_gpus": 1,
"num_workers": 1,
"monitor": True,
"buffer_size": 2000,
},
)