sb3

/

ppo-Walker2d-v3

like 0

Follow

Stable-Baselines3 32

Hello,

I am attempting to use the ppo-Walker2d-v3.zip model, but am unable to get near the reported mean return of 3571. Instead, I consistently get a return of ~1 for episodes of 1000 steps. The snippet below illustrates how I am setting up the environment, perhaps the issue is here?

from sb3_contrib.common.wrappers import TimeFeatureWrapper
from gym.wrappers import NormalizeObservation
from stable_baselines3 import PPO
from huggingface_sb3 import load_from_hub
import gym

checkpoint = load_from_hub(
    repo_id="sb3/ppo-Walker2d-v3",
    filename="ppo-Walker2d-v3.zip",
)
custom_objects = {
    "learning_rate": 0.0,
    "lr_schedule": lambda _: 0.0,
    "clip_range": lambda _: 0.0,
}
expert_model = PPO.load(checkpoint,custom_objects=custom_objects)

env_name = "Walker2d-v3"
venv = gym.make(env_name)
venv = NormalizeObservation(venv)
venv = TimeFeatureWrapper(venv)