Duplicate from huggingface-projects/Deep-Reinforcement-Learning-Leaderboard

.gitattributes

@@ -0,0 +1,27 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zstandard filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1 @@
+__pycache__/*

README.md

@@ -0,0 +1,13 @@
+---
+title: Deep Reinforcement Learning Leaderboard
+emoji: 🚀
+colorFrom: green
+colorTo: indigo
+sdk: gradio
+sdk_version: 3.0.20
+app_file: app.py
+pinned: false
+duplicated_from: huggingface-projects/Deep-Reinforcement-Learning-Leaderboard
+---
+
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference

app.css

@@ -0,0 +1,37 @@
+
+.infoPoint h1 {
+    font-size: 30px;
+    text-decoration: bold;
+    
+    }
+    
+a {
+    text-decoration: underline; 
+    color: #1f3b54 ;
+}
+
+table {
+  
+    margin: 25px 0;
+    font-size: 0.9em;
+    font-family: sans-serif;
+    min-width: 400px;
+    box-shadow: 0 0 20px rgba(0, 0, 0, 0.15);
+}
+
+table th,
+table td {
+    padding: 12px 15px;
+}
+
+tr {
+text-align: left;
+}
+thead tr {
+text-align: left;
+}
+
+.flex
+{
+    overflow:auto;
+}

app.py

@@ -0,0 +1,238 @@
+import requests
+import pandas as pd
+from tqdm.auto import tqdm
+from utils import *
+import gradio as gr
+
+from huggingface_hub import HfApi, hf_hub_download
+from huggingface_hub.repocard import metadata_load
+
+class DeepRL_Leaderboard:
+    def __init__(self) -> None:
+        self.leaderboard= {} 
+
+    def add_leaderboard(self,id=None, title=None):
+        if id is not None and title is not None:
+            id = id.strip()
+            title = title.strip()
+            self.leaderboard.update({id:{'title':title,'data':get_data_per_env(id)}})
+            
+    def get_data(self):
+        return self.leaderboard
+
+    def get_ids(self):
+        return list(self.leaderboard.keys())
+
+# CSS file for the
+with open('app.css','r') as f:
+    BLOCK_CSS = f.read() 
+
+LOADED_MODEL_IDS = {}
+LOADED_MODEL_METADATA = {}
+
+def get_data(rl_env):
+    global LOADED_MODEL_IDS ,LOADED_MODEL_METADATA
+    data = []
+    model_ids = get_model_ids(rl_env)
+    LOADED_MODEL_IDS[rl_env]=model_ids
+
+    for model_id in tqdm(model_ids):
+        meta = get_metadata(model_id)
+        LOADED_MODEL_METADATA[model_id] = meta if meta is not None else ''
+        if meta is None:
+            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
+        data.append(row)
+    return pd.DataFrame.from_records(data)
+
+def get_data_per_env(rl_env):
+    dataframe = get_data(rl_env)
+    dataframe = dataframe.fillna("")
+
+    if not dataframe.empty:
+        # turn the model ids into clickable links
+        dataframe["User"] = dataframe["User"].apply(make_clickable_user)
+        dataframe["Model"] = dataframe["Model"].apply(make_clickable_model)
+        dataframe = dataframe.sort_values(by=['Results'], ascending=False)
+        if not 'Ranking' in dataframe.columns:
+            dataframe.insert(0, 'Ranking', [i for i in range(1,len(dataframe)+1)])
+        else:
+           dataframe['Ranking'] =   [i for i in range(1,len(dataframe)+1)]   
+        table_html = dataframe.to_html(escape=False, index=False,justify = 'left')
+        return table_html,dataframe,dataframe.empty
+    else: 
+        html = """<div style="color: green">
+                <p> ⌛ Please wait. Results will be out soon... </p>
+                </div>
+               """
+        return html,dataframe,dataframe.empty   
+
+
+
+rl_leaderboard = DeepRL_Leaderboard()
+rl_leaderboard.add_leaderboard('CartPole-v1','The Cartpole-v1 Leaderboard')
+rl_leaderboard.add_leaderboard('LunarLander-v2',"The Lunar Lander 🌕 Leaderboard")
+rl_leaderboard.add_leaderboard('FrozenLake-v1-4x4-no_slippery','The FrozenLake-v1-4x4-no_slippery Leaderboard')
+rl_leaderboard.add_leaderboard('FrozenLake-v1-8x8-no_slippery','The FrozenLake-v1-8x8-no_slippery Leaderboard')
+rl_leaderboard.add_leaderboard('FrozenLake-v1-4x4','The FrozenLake-v1-4x4 Leaderboard')
+rl_leaderboard.add_leaderboard('FrozenLake-v1-8x8','The FrozenLake-v1-8x8 Leaderboard')
+rl_leaderboard.add_leaderboard('Taxi-v3','The Taxi-v3🚖 Leaderboard')
+rl_leaderboard.add_leaderboard('CarRacing-v0'," The Car Racing 🏎️ Leaderboard")
+rl_leaderboard.add_leaderboard('MountainCar-v0',"The Mountain Car ⛰️ 🚗 Leaderboard")
+rl_leaderboard.add_leaderboard('BipedalWalker-v3',"The BipedalWalker Leaderboard")
+rl_leaderboard.add_leaderboard('SpaceInvadersNoFrameskip-v4','The SpaceInvadersNoFrameskip-v4 Leaderboard')
+rl_leaderboard.add_leaderboard('Pixelcopter-PLE-v0','The Pixelcopter-PLE-v0 🚁 Leaderboard')
+rl_leaderboard.add_leaderboard('Pong-PLE-v0','The Pong-PLE-v0 🎾 Leaderboard')
+rl_leaderboard.add_leaderboard('Walker2DBulletEnv-v0','The Walker2DBulletEnv-v0 🤖 Leaderboard')
+rl_leaderboard.add_leaderboard('AntBulletEnv-v0','The AntBulletEnv-v0 🕸️ Leaderboard')
+rl_leaderboard.add_leaderboard('HalfCheetahBulletEnv-v0','The HalfCheetahBulletEnv-v0 🤖 Leaderboard')
+RL_ENVS = rl_leaderboard.get_ids()
+RL_DETAILS = rl_leaderboard.get_data()
+
+
+def update_data(rl_env):
+    global LOADED_MODEL_IDS,LOADED_MODEL_METADATA
+    data = []
+    model_ids = [x for x in get_model_ids(rl_env)] #if x not in LOADED_MODEL_IDS[rl_env]] # For now let's update all
+    
+    LOADED_MODEL_IDS[rl_env]+=model_ids
+
+    for model_id in tqdm(model_ids):
+        meta = get_metadata(model_id)
+        LOADED_MODEL_METADATA[model_id] = meta if meta is not None else ''
+        if meta is None:
+            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
+        data.append(row)
+    return pd.DataFrame.from_records(data)
+
+
+def update_data_per_env(rl_env):
+    global RL_DETAILS
+
+    _,old_dataframe,_ = RL_DETAILS[rl_env]['data']
+    new_dataframe = update_data(rl_env)
+
+    new_dataframe = new_dataframe.fillna("")
+    if not new_dataframe.empty:
+        new_dataframe["User"] = new_dataframe["User"].apply(make_clickable_user)
+        new_dataframe["Model"] = new_dataframe["Model"].apply(make_clickable_model)
+
+    dataframe = pd.concat([old_dataframe,new_dataframe])
+
+    if not dataframe.empty:
+       
+        dataframe = dataframe.sort_values(by=['Results'], ascending=False)
+        if not 'Ranking' in dataframe.columns:
+            dataframe.insert(0, 'Ranking', [i for i in range(1,len(dataframe)+1)])
+        else:
+           dataframe['Ranking'] =   [i for i in range(1,len(dataframe)+1)]   
+        table_html = dataframe.to_html(escape=False, index=False,justify = 'left')
+        return table_html,dataframe,dataframe.empty
+    else: 
+        html = """<div style="color: green">
+                <p> ⌛ Please wait. Results will be out soon... </p>
+                </div>
+               """
+        return html,dataframe,dataframe.empty   
+
+
+def get_info_display(dataframe,env_name,name_leaderboard,is_empty):
+    if not is_empty:
+        markdown = """
+        <div class='infoPoint'>
+        <h1> {name_leaderboard} </h1>
+        <br>
+        <p> This is a leaderboard of <b>{len_dataframe}</b> agents, from <b>{num_unique_users}</b> unique users, playing {env_name} 👩‍🚀. </p>
+        <br>
+        <p> We use <b>lower bound result to sort the models: mean_reward - std_reward.</b> </p>
+        <br>    
+        <p> You can click on the model's name to be redirected to its model card which includes documentation. </p>
+        <br>
+        <p> You want to try to train your agents? <a href="http://eepurl.com/h1pElX" target="_blank">Sign up to the Hugging Face free Deep Reinforcement Learning Class 🤗 </a>.
+        </p>
+        <br>
+        <p> You want to compare two agents? <a href="https://huggingface.co/spaces/ThomasSimonini/Compare-Reinforcement-Learning-Agents" target="_blank">It's possible using this Spaces demo 👀 </a>.
+        </p>
+        </div>
+        """.format(len_dataframe = len(dataframe),env_name = env_name,name_leaderboard = name_leaderboard,num_unique_users = len(set(dataframe['User'].values)))
+
+    else:
+        markdown = """
+        <div class='infoPoint'>
+        <h1> {name_leaderboard} </h1>
+        <br>
+        </div>                  
+        """.format(name_leaderboard =  name_leaderboard)
+    return markdown    
+
+def reload_all_data():
+
+    global RL_DETAILS,RL_ENVS
+
+    for rl_env in RL_ENVS:
+        RL_DETAILS[rl_env]['data'] = update_data_per_env(rl_env)
+
+    html = """<div style="color: green">
+                <p> ✅ Leaderboard updated! </p>
+                </div>
+               """    
+    return html            
+
+
+def reload_leaderboard(rl_env):
+    global RL_DETAILS
+ 
+    data_html,data_dataframe,is_empty = RL_DETAILS[rl_env]['data'] 
+
+    markdown = get_info_display(data_dataframe,rl_env,RL_DETAILS[rl_env]['title'],is_empty)            
+    
+    return markdown,data_html     
+            
+
+           
+
+
+
+block = gr.Blocks(css=BLOCK_CSS)
+with block:
+    notification = gr.HTML("""<div style="color: green">
+                <p> ⌛ Updating leaderboard... </p>
+                </div>
+               """)
+    block.load(reload_all_data,[],[notification])
+    
+    with gr.Tabs():
+        for rl_env in RL_ENVS:
+            with gr.TabItem(rl_env) as rl_tab:
+                data_html,data_dataframe,is_empty = RL_DETAILS[rl_env]['data'] 
+                markdown = get_info_display(data_dataframe,rl_env,RL_DETAILS[rl_env]['title'],is_empty)            
+                env_state =gr.Variable(value=f'\"{rl_env}\"')  
+                output_markdown = gr.HTML(markdown)
+
+                output_html = gr.HTML(data_html)
+
+                rl_tab.select(reload_leaderboard,inputs=[env_state],outputs=[output_markdown,output_html])
+
+block.launch()    

utils.py

@@ -0,0 +1,68 @@
+import pandas as pd
+import requests
+from tqdm.auto import tqdm
+from huggingface_hub import HfApi, hf_hub_download
+from huggingface_hub.repocard import metadata_load
+
+
+# Based on Omar Sanseviero work
+# Make model clickable link
+def make_clickable_model(model_name):
+    # remove user from model name
+    model_name_show = ' '.join(model_name.split('/')[1:])
+
+    link = "https://huggingface.co/" + model_name
+    return f'<a target="_blank" href="{link}">{model_name_show}</a>'
+
+# Make user clickable link
+def make_clickable_user(user_id):
+    link = "https://huggingface.co/" + user_id
+    return f'<a  target="_blank" href="{link}">{user_id}</a>'
+    
+
+
+def get_model_ids(rl_env):
+    api = HfApi()
+    models = api.list_models(filter=rl_env)
+    model_ids = [x.modelId for x in models]
+    return model_ids
+    
+def get_metadata(model_id):
+    try:
+        readme_path = hf_hub_download(model_id, filename="README.md")
+        return metadata_load(readme_path)
+    except requests.exceptions.HTTPError:
+        # 404 README.md not found
+        return None
+        
+def parse_metrics_accuracy(meta):
+    if "model-index" not in meta:
+        return None
+    result = meta["model-index"][0]["results"]
+    metrics = result[0]["metrics"]
+    accuracy = metrics[0]["value"]
+    return accuracy
+
+# We keep the worst case episode
+def parse_rewards(accuracy):
+    default_std = -1000
+    default_reward=-1000
+    if accuracy !=  None:
+        accuracy = str(accuracy)
+        parsed =  accuracy.split(' +/- ')
+        if len(parsed)>1:
+            mean_reward = float(parsed[0])
+            std_reward =  float(parsed[1])
+        elif len(parsed)==1: #only mean reward   
+            mean_reward = float(parsed[0])
+            std_reward =  float(0)
+             
+        else: 
+            mean_reward = float(default_std)
+            std_reward = float(default_reward)
+
+    else:
+        mean_reward = float(default_std)
+        std_reward = float(default_reward)
+    return mean_reward, std_reward
+