Views
No views yet
1from stable_baselines3.common.vec_env import DummyVecEnv, VecNormalize
2
3# Load the saved statistics
4eval_env = DummyVecEnv([lambda: gym.make("PandaReachDense-v3")])
5eval_env = VecNormalize.load("vec_normalize.pkl", eval_env)
6
7# We need to override the render_mode
8eval_env.render_mode = "rgb_array"
9
10# do not update them at test time
11eval_env.training = False
12# reward normalization is not needed at test time
13eval_env.norm_reward = False
14
15# Load the agent
16model = A2C.load("a2c-PandaReachDense-v3")
17
18mean_reward, std_reward = evaluate_policy(model, eval_env)
19
20print(f"Mean reward = {mean_reward:.2f} +/- {std_reward:.2f}")
21
22...