add lot of envs

app.py

@@ -9,9 +9,10 @@ from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import HfApi, hf_hub_download
 
 from src.backend import backend_routine
-from src.css_html_js import dark_mode_gradio_js
+
 from src.logging import configure_root_logger, setup_logger
 
+
 configure_root_logger()
 logger = setup_logger(__name__)
 
@@ -19,20 +20,101 @@ API = HfApi(token=os.environ.get("TOKEN"))
 RESULTS_REPO = f"open-rl-leaderboard/results"
 ALL_ENV_IDS = {
     "Atari": [
-        "BeamRiderNoFrameskip-v4",
-        "BreakoutNoFrameskip-v4",
+        "Adventure",
+        "AirRaid",
+        "Alien",
+        "Amidar",
+        "Assault",
+        "Asterix",
+        "Asteroids",
+        "Atlantis",
+        "BankHeist",
+        "BattleZone",
+        "BeamRider",
+        "Berzerk",
+        "Bowling",
+        "Boxing",
+        "Breakout",
+        "Carnival",
+        "Centipede",
+        "ChopperCommand",
+        "CrazyClimber",
+        "Defender",
+        "DemonAttack",
+        "DoubleDunk",
+        "ElevatorAction",
+        "Enduro",
+        "FishingDerby",
+        "Freeway",
+        "Frostbite",
+        "Gopher",
+        "Gravitar",
+        "Hero",
+        "IceHockey",
+        "Jamesbond",
+        "JourneyEscape",
+        "Kangaroo",
+        "Krull",
+        "KungFuMaster",
+        "MontezumaRevenge",
+        "MsPacman",
+        "NameThisGame",
+        "Phoenix",
+        "Pitfall",
+        "Pong",
+        "Pooyan",
+        "PrivateEye",
+        "Qbert",
+        "Riverraid",
+        "RoadRunner",
+        "Robotank",
+        "Seaquest",
+        "Skiing",
+        "Solaris",
+        "SpaceInvaders",
+        "StarGunner",
+        "Tennis",
+        "TimePilot",
+        "Tutankham",
+        "UpNDown",
+        "Venture",
+        "VideoPinball",
+        "WizardOfWor",
+        "YarsRevenge",
+        "Zaxxon",
     ],
     "Box2D": [
         "LunarLander-v2",
+        "LunarLanderContinuous-v2",
         "BipedalWalker-v3",
+        "BipedalWalkerHardcore-v3",
+        "CarRacing-v2",
+    ],
+    "Toy text": [
+        "Blackjack-v1",
+        "FrozenLake-v1",
+        "FrozenLake8x8-v1",
+        "CliffWalking-v0",
     ],
     "Classic control": [
+        "Acrobot-v1",
         "CartPole-v1",
         "MountainCar-v0",
+        "MountainCarContinuous-v0",
+        "Pendulum-v1",
     ],
     "MuJoCo": [
-        "Hopper-v4",
+        "Reacher-v4",
+        "Pusher-v4",
+        "InvertedPendulum-v4",
+        "InvertedDoublePendulum-v4",
         "HalfCheetah-v4",
+        "Hopper-v4",
+        "Swimmer-v4",
+        "Walker2d-v4",
+        "Ant-v4",
+        "Humanoid-v4",
+        "HumanoidStandup-v4",
     ],
 }
 
@@ -77,12 +159,14 @@ The Open RL Leaderboard is a community-driven benchmark for reinforcement learni
 
 def select_env(df: pd.DataFrame, env_id: str):
     df = df[df["env_id"] == env_id]
-
-    # Add the ranking
     df = df.sort_values("mean_episodic_return", ascending=False)
     df["ranking"] = np.arange(1, len(df) + 1)
+    return df
 
+
+def format_df(df: pd.DataFrame):
     # Add hyperlinks
+    df = df.copy()
     for index, row in df.iterrows():
         user_id = row["user_id"]
         model_id = row["model_id"]
@@ -105,26 +189,42 @@ with gr.Blocks() as demo:
                     for env_id in env_ids:
                         with gr.TabItem(env_id):
                             with gr.Row(equal_height=False):
+                                if env_domain == "Atari":
+                                    env_id = f"{env_id}NoFrameskip-v4"
+                                env_df = select_env(df, env_id)
                                 gr.components.Dataframe(
-                                    value=select_env(df, env_id),
+                                    value=format_df(env_df),
                                     headers=["🏆 Ranking", "🧑 User", "🤖 Model id", "📊 Mean episodic return"],
                                     datatype=["number", "markdown", "markdown", "number"],
                                     row_count=(10, "fixed"),
                                     scale=3,
                                 )
-                                gr.Video(
-                                    "https://huggingface.co/qgallouedec/MsPacmanNoFrameskip-v4-dqn_atari-seed1/resolve/main/replay.mp4",
-                                    autoplay=True,
-                                    scale=1,
-                                    min_width=50,
-                                )
+                                # Get the best model and
+                                if not env_df.empty:
+                                    user_id = env_df.iloc[0]["user_id"]
+                                    model_id = env_df.iloc[0]["model_id"]
+                                    video_path = hf_hub_download(
+                                        repo_id=f"{user_id}/{model_id}",
+                                        filename="replay.mp4",
+                                        revision="main",
+                                        repo_type="model",
+                                    )
+                                    video = gr.PlayableVideo(
+                                        video_path,
+                                        autoplay=True,
+                                        scale=1,
+                                        min_width=50,
+                                        show_download_button=False,
+                                        label=model_id,
+                                    )
+                                    # Doesn't loop for the moment, see https://github.com/gradio-app/gradio/issues/7689
 
         with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
             gr.Markdown(ABOUT_TEXT)
 
 
 scheduler = BackgroundScheduler()
-scheduler.add_job(func=backend_routine, trigger="interval", seconds=0.5 * 60, max_instances=1)
+scheduler.add_job(func=backend_routine, trigger="interval", seconds=10 * 60, max_instances=1)
 scheduler.start()