Commit
•
2c1d722
1
Parent(s):
5adbf42
Update README.md
Browse filesCode implementation to evaluate the PPO model in readme file
README.md
CHANGED
@@ -26,12 +26,41 @@ This is a trained model of a **PPO** agent playing **LunarLander-v2**
|
|
26 |
using the [stable-baselines3 library](https://github.com/DLR-RM/stable-baselines3).
|
27 |
|
28 |
## Usage (with Stable-baselines3)
|
29 |
-
TODO: Add your code
|
30 |
-
|
31 |
|
32 |
```python
|
33 |
from stable_baselines3 import ...
|
34 |
from huggingface_sb3 import load_from_hub
|
35 |
|
36 |
-
...
|
37 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
using the [stable-baselines3 library](https://github.com/DLR-RM/stable-baselines3).
|
27 |
|
28 |
## Usage (with Stable-baselines3)
|
|
|
|
|
29 |
|
30 |
```python
|
31 |
from stable_baselines3 import ...
|
32 |
from huggingface_sb3 import load_from_hub
|
33 |
|
|
|
34 |
```
|
35 |
+
|
36 |
+
Use the model like this
|
37 |
+
|
38 |
+
```python
|
39 |
+
import gym
|
40 |
+
|
41 |
+
from huggingface_sb3 import load_from_hub
|
42 |
+
from stable_baselines3 import PPO
|
43 |
+
from stable_baselines3.common.evaluation import evaluate_policy
|
44 |
+
|
45 |
+
# Retrieve the model from the hub
|
46 |
+
## repo_id = id of the model repository from the Hugging Face Hub (repo_id = {organization}/{repo_name})
|
47 |
+
## filename = name of the model zip file from the repository
|
48 |
+
checkpoint = load_from_hub(repo_id="ThomasSimonini/ppo-LunarLander-v2", filename="ppo-LunarLander-v2.zip")
|
49 |
+
model = PPO.load(checkpoint)
|
50 |
+
|
51 |
+
# Evaluate the agent
|
52 |
+
eval_env = gym.make('LunarLander-v2')
|
53 |
+
mean_reward, std_reward = evaluate_policy(model, eval_env, n_eval_episodes=10, deterministic=True)
|
54 |
+
print(f"mean_reward={mean_reward:.2f} +/- {std_reward}")
|
55 |
+
|
56 |
+
# Watch the agent play
|
57 |
+
obs = eval_env.reset()
|
58 |
+
for i in range(1000):
|
59 |
+
action, _state = model.predict(obs)
|
60 |
+
obs, reward, done, info = eval_env.step(action)
|
61 |
+
eval_env.render()
|
62 |
+
if done:
|
63 |
+
obs = eval_env.reset()
|
64 |
+
eval_env.close()
|
65 |
+
|
66 |
+
```
|