chore: add more required envs and more

app.py

@@ -24,111 +24,73 @@ rl_envs = [
 {
 "rl_env_beautiful": "LunarLander-v2 🚀",
 "rl_env": "LunarLander-v2",
-"video_link": "",
-"global": None
-},    
-{
-"rl_env_beautiful": "FrozenLake-v1-4x4-no_slippery ❄️",
-"rl_env": "FrozenLake-v1-4x4-no_slippery",
-"video_link": "",
-"global": None
+"unit": "Unit 1",
+"library": "stable-baselines3",
+"min_result": 200,
 },
 {
 "rl_env_beautiful": "Taxi-v3 🚖",
 "rl_env": "Taxi-v3",
-"video_link": "",
-"global": None
+"unit": "Unit 2",
+"library": "q-learning",
+"min_result": 4,
 },
 {
 "rl_env_beautiful": "SpaceInvadersNoFrameskip-v4 👾",
 "rl_env": "SpaceInvadersNoFrameskip-v4",
-"video_link": "",
-"global": None
+"unit": "Unit 3",
+"library": "stable-baselines3",
+"min_result": 200,
 },
 {
 "rl_env_beautiful": "CartPole-v1",
 "rl_env": "CartPole-v1",
-"video_link": "https://huggingface.co/sb3/ppo-CartPole-v1/resolve/main/replay.mp4",
-"global": None
+"unit": "Unit 4",
+"library": "reinforce",
+"min_result": 350,
 },
 {
 "rl_env_beautiful": "Pixelcopter-PLE-v0",
 "rl_env": "Pixelcopter-PLE-v0",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "CarRacing-v0 🏎️",
-"rl_env": "CarRacing-v0",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "CarRacing-v2 🏎️",
-"rl_env": "CarRacing-v2",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "MountainCar-v0 ⛰️",
-"rl_env": "MountainCar-v0",
-"video_link": "",
-"global": None
+"unit": "Unit 4",
+"library": "reinforce",
+"min_result": 5,
 },
 {
-"rl_env_beautiful": "PongNoFrameskip-v4 🎾",
-"rl_env": "PongNoFrameskip-v4",
-"video_link": "",
-"global": None
+"rl_env_beautiful": "ML-Agents Snowball Target ❄️",
+"rl_env": "ML-Agents-SnowballTarget",
+"unit": "Unit 5",
+"library": "ml-agents",
+"min_result": -100,
 },
 {
-"rl_env_beautiful": "BreakoutNoFrameskip-v4 🧱",
-"rl_env": "BreakoutNoFrameskip-v4",
-"video_link": "",
-"global": None
+"rl_env_beautiful": "ML-Agents Pyramids 🏔️",
+"rl_env": "ML-Agents-Pyramids",
+"unit": "Unit 5",
+"library": "ml-agents",
+"min_result": -100,
 },
 {
-"rl_env_beautiful": "QbertNoFrameskip-v4 🐦",
-"rl_env": "QbertNoFrameskip-v4",
-"video_link": "",
-"global": None
+"rl_env_beautiful": "Panda Reach Dense 🤖",
+"rl_env": "PandaReachDense",
+"unit": "Unit 6",
+"library": "stable-baselines3",
+"min_result": -3.5,
 },
 {
-"rl_env_beautiful": "BipedalWalker-v3",
-"rl_env": "BipedalWalker-v3",
-"video_link": "",
-"global": None
+"rl_env_beautiful": "ML-Agents Soccer Twos ⚽",
+"rl_env": "ML-Agents-SoccerTwos",
+"unit": "Unit 7",
+"library": "ml-agents",
+"min_result": -100,
 },
 {
-"rl_env_beautiful": "Walker2DBulletEnv-v0",
-"rl_env": "Walker2DBulletEnv-v0",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "AntBulletEnv-v0",
-"rl_env": "AntBulletEnv-v0",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "HalfCheetahBulletEnv-v0",
-"rl_env": "HalfCheetahBulletEnv-v0",
-"video_link": "",
-"global": None
-},
-{
-"rl_env_beautiful": "PandaReachDense-v2",
-"rl_env": "PandaReachDense-v2",
-"video_link": "",
-"global": None
-},  
-{
-"rl_env_beautiful": "PandaReachDense-v3",
-"rl_env": "PandaReachDense-v3",
-"video_link": "",
-"global": None
-},
+"rl_env_beautiful": "Doom Health Gathering Supreme",
+"rl_env": "doom_health_gathering_supreme",
+"unit": "Unit 8 PII",
+"library": "sample-factory",
+"min_result": 5,
+}
 ]
 
 def restart():
@@ -174,42 +136,94 @@ def parse_rewards(accuracy):
     return mean_reward, std_reward
 
 
-def get_model_ids(rl_env):
+def get_user_models(hf_username, env_tag, lib_tag):
+    """
+    List the Reinforcement Learning models
+    from user given environment and lib
+    :param hf_username: User HF username
+    :param env_tag: Environment tag
+    :param lib_tag: Library tag
+    """
     api = HfApi()
-    models = api.list_models(filter=rl_env)
-    model_ids = [x.modelId for x in models]
-    return model_ids
+    models = api.list_models(author=hf_username, filter=["reinforcement-learning", env_tag, lib_tag])
 
-def filter_students(model_ids):
-    filtered = []
-    for model_id in model_ids:
-        user_id = model_id.split('/')[0]
-        if user_id in STUDENTS_SET:
-            filtered.append(model_id)
-    return filtered
+    user_model_ids = [(x.modelId, (x.created_at or x.last_modified)) for x in models]
+    return user_model_ids
+
+
+def get_user_sf_models(hf_username, env_tag, lib_tag):
+    models_sf = []
+    models = api.list_models(author=hf_username, filter=["reinforcement-learning", lib_tag])
+
+    user_model_ids = [(x.modelId, (x.created_at or x.last_modified)) for x in models]
+
+    for model, last_updated in user_model_ids:
+        meta = get_metadata(model)
+        if meta is None:
+            continue
+        result = meta["model-index"][0]["results"][0]["dataset"]["name"]
+        if result == env_tag:
+            models_sf.append((model, last_updated))
+            
+    return models_sf
+
+
+def calculate_best_result(user_model_ids):
+  """
+  Calculate the best results of a unit
+  best_result = mean_reward - std_reward
+  :param user_model_ids: RL models of a user
+  """
+  best_result = -1000
+  best_model_id = ""
+  best_last_updated = None
+  for model, last_updated in user_model_ids:
+    meta = get_metadata(model)
+    if meta is None:
+      continue
+    accuracy = parse_metrics_accuracy(meta)
+    mean_reward, std_reward = parse_rewards(accuracy)
+    result = mean_reward - std_reward
+    if result > best_result:
+      best_result = result
+      best_model_id = model
+      best_last_updated = last_updated
+      
+  return best_result, best_model_id, best_last_updated
+
+def get_model_ids(hf_username, rl_env):
+    if rl_env["rl_env"] == "PandaReachDense":
+        # Since Unit 6 can use PandaReachDense-v2 or v3
+        user_models = get_user_models(hf_username, "PandaReachDense-v3", rl_env["library"])
+        if len(user_models) == 0:
+            user_models = get_user_models(hf_username, "PandaReachDense-v2", rl_env["library"])
+    elif rl_env["rl_env"] != "doom_health_gathering_supreme":
+        user_models = get_user_models(hf_username, rl_env["rl_env"], rl_env["library"])
+    else:
+        user_models = get_user_sf_models(hf_username, rl_env["rl_env"], rl_env["library"])
+    
+    # Calculate the best result and get the best_model_id
+    best_result, best_model_id, best_last_updated = calculate_best_result(user_models)
+    passed = best_result >= rl_env["min_result"]
+    return best_model_id, best_result, best_last_updated, passed
 
 
 def update_leaderboard_dataset(rl_env, path):
     # Get model ids associated with rl_env
-    model_ids = get_model_ids(rl_env)
-    model_ids = filter_students(model_ids)
+    model_info = []
+    for user_id in STUDENTS_SET:
+        model_info.append(get_model_ids(user_id, rl_env))
     data = []
-    for model_id in model_ids:
-        meta = get_metadata(model_id)
-        #LOADED_MODEL_METADATA[model_id] = meta if meta is not None else ''
-        if meta is None:
+    for model_id, result, updated, passed in model_info:
+        if model_id is None or model_id == "":
             continue
         user_id = model_id.split('/')[0]
         row = {}
         row["User"] = user_id
         row["Model"] = model_id
-        accuracy = parse_metrics_accuracy(meta)
-        mean_reward, std_reward = parse_rewards(accuracy)
-        mean_reward = mean_reward if not pd.isna(mean_reward) else 0
-        std_reward = std_reward if not pd.isna(std_reward) else 0
-        row["Results"] = mean_reward - std_reward
-        row["Mean Reward"] = mean_reward
-        row["Std Reward"] = std_reward
+        row["Result"] = result
+        row["Submitted"] = updated
+        row["Passed"] = passed
         data.append(row)
     
     if not data:
@@ -217,7 +231,7 @@ def update_leaderboard_dataset(rl_env, path):
 
     ranked_dataframe = rank_dataframe(pd.DataFrame.from_records(data))
     new_history = ranked_dataframe
-    file_path = path + "/" + rl_env + ".csv"
+    file_path = path + "/" + rl_env["rl_env"] + ".csv"
     new_history.to_csv(file_path, index=False)
 
 def download_leaderboard_dataset():
@@ -231,7 +245,7 @@ def get_data(rl_env, path) -> pd.DataFrame:
     """
     csv_path = path + "/" + rl_env + ".csv"
     if not os.path.exists(csv_path):
-        return pd.DataFrame(columns=['Ranking', 'User', 'Model', 'Results', 'Mean Reward', 'Std Reward'])
+        return pd.DataFrame(columns=['Ranking', 'User', 'Model', 'Result', 'Submitted', 'Passed'])
 
     data = pd.read_csv(csv_path)
 
@@ -255,8 +269,8 @@ def get_data_no_html(rl_env, path) -> pd.DataFrame:
     
 def rank_dataframe(dataframe):
     if dataframe.empty:
-        return pd.DataFrame(columns=['User', 'Model', 'Results', 'Mean Reward', 'Std Reward'])
-    dataframe = dataframe.sort_values(by=['Results', 'User', 'Model'], ascending=False)
+        return pd.DataFrame(columns=['User', 'Model', 'Result', 'Passed'])
+    dataframe = dataframe.sort_values(by=['Result'], ascending=False)
     if not 'Ranking' in dataframe.columns:
         dataframe.insert(0, 'Ranking', [i for i in range(1,len(dataframe)+1)])
     else:
@@ -268,7 +282,7 @@ def run_update_dataset():
     path_ = download_leaderboard_dataset()
     for i in range(0, len(rl_envs)):
         rl_env = rl_envs[i]
-        update_leaderboard_dataset(rl_env["rl_env"], path_)
+        update_leaderboard_dataset(rl_env, path_)
 
     api.upload_folder(
     folder_path=path_,
@@ -297,14 +311,15 @@ with block:
         with gr.TabItem(rl_env["rl_env_beautiful"]) as rl_tab:
             with gr.Row():
                 markdown = """
-                    # {name_leaderboard}
+                    # {unit}
+                    ## {name_leaderboard}
                     
-                    """.format(name_leaderboard = rl_env["rl_env_beautiful"], video_link = rl_env["video_link"])
+                    """.format(name_leaderboard = rl_env["rl_env_beautiful"], unit=rl_env["unit"])
                 gr.Markdown(markdown)
                 
             
             with gr.Row():
-                gr_dataframe = gr.components.Dataframe(value=get_data(rl_env["rl_env"], path_), headers=["Ranking 🏆", "User 🤗", "Model id 🤖", "Results", "Mean Reward", "Std Reward"], datatype=["number", "markdown", "markdown", "number", "number", "number"],  row_count=(15, 'dynamic'))
+                gr_dataframe = gr.components.Dataframe(value=get_data(rl_env["rl_env"], path_), headers=["Ranking 🏆", "User 🤗", "Model id 🤖", "Result",  "Submitted", "Passed"], datatype=["number", "markdown", "markdown", "number", "date", "bool"],  row_count=(15, 'dynamic'))
     """
     block.load(
         download_leaderboard_dataset,

requirements.txt

@@ -1,4 +1,4 @@
-APScheduler==3.10.1
+APScheduler==3.11.1
 gradio==5.49.1
 httpx>=0.24.1
 tqdm