Enduro-v5-PPO / config.json
lucasschott's picture
improved agent
4e15a01
Raw History Blame Contribute Delete
461 Bytes
{
"total_timesteps": 40000000,
"policy": "MlpPolicy",
"policy_kwargs" : {
"features_extractor_class" : "NatureCNN",
"activation_fn" : "nn.ReLU",
"net_arch" : "dict(vf=[512, 256], pi=[256, 128])"
},
"learning_rate" : "linear_schedule(3e-4,1e-6)",
"n_steps" : 256,
"batch_size" : 1024,
"n_epochs" : 4,
"gamma" : 0.99,
"gae_lambda" : 0.95,
"clip_range" : 0.1,
"vf_coef" : 0.5,
"ent_coef" : 0.01
}