Leaderboard: names; percentage; sort by elo&votes

app/leaderboard.py

@@ -12,40 +12,51 @@ def get_leaderboard(reveal_prelim = False):
 
     conn = get_db()
     cursor = conn.cursor()
-    sql = 'SELECT name, upvote, downvote, name AS orig_name FROM model'
+    sql = 'SELECT Name, Upvote, Downvote, name AS OrigName FROM model'
     prelim_votes = 300
-    if not reveal_prelim: sql += ' WHERE (upvote + downvote) > '+ str(prelim_votes)
+    if not reveal_prelim: sql += ' WHERE (Upvote + Downvote) > '+ str(prelim_votes)
     cursor.execute(sql)
     data = cursor.fetchall()
-    df = pd.DataFrame(data, columns=['name', 'upvote', 'downvote', 'orig_name'])
-    # df['license'] = df['name'].map(model_license)
-    df['name'] = df['name'].replace(model_names)
+    df = pd.DataFrame(data, columns=['Name', 'Upvote', 'Downvote', 'OrigName'])
+    # df['License'] = df['Name'].map(model_license)
+    df['Name'] = df['Name'].replace(model_names)
     for i in range(len(df)):
-        df.loc[i, "name"] = make_link_to_space(df['name'][i], True)
+        df.loc[i, "Name"] = make_link_to_space(df['Name'][i], True)
 
     # Calculate total votes and win rate
-    df['votes'] = df['upvote'] + df['downvote']
-    df['win_rate'] = (df['upvote'] / df['votes'] * 100).round(1)
-    # df['score'] = round((df['upvote'] / df['votes']) * 100, 2) # Percentage score
+    df['Votes'] = df['Upvote'] + df['Downvote']
+    df['Win Rate'] = (df['Upvote'] / df['Votes'] * 100).round(1)
+    # convert to string
+    df['Win Rate'] = df['Win Rate'].astype(str) + '%'
+
+    # df['Score'] = round((df['Upvote'] / df['Votes']) * 100, 2) # Percentage score
 
     ## ELO SCORE
-    df['elo'] = 1200
-    df['elo_diff'] = ""
+    df['Elo'] = 1200
+    df['Elo Diff'] = ""
     for i in range(len(df)):
         for j in range(len(df)):
             if i != j:
                 try:
-                    expected_a = 1 / (1 + 10 ** ((df['elo'].iloc[j] - df['elo'].iloc[i]) / 400))
-                    expected_b = 1 / (1 + 10 ** ((df['elo'].iloc[i] - df['elo'].iloc[j]) / 400))
-                    actual_a = df['upvote'].iloc[i] / df['votes'].iloc[i] if df['votes'].iloc[i] > 0 else 0.5
-                    actual_b = df['upvote'].iloc[j] / df['votes'].iloc[j] if df['votes'].iloc[j] > 0 else 0.5
-                    df.at[i, 'elo'] += round(32 * (actual_a - expected_a))
-                    df.at[j, 'elo'] += round(32 * (actual_b - expected_b))
+                    expected_a = 1 / (1 + 10 ** ((df['Elo'].iloc[j] - df['Elo'].iloc[i]) / 400))
+                    expected_b = 1 / (1 + 10 ** ((df['Elo'].iloc[i] - df['Elo'].iloc[j]) / 400))
+                    actual_a = df['Upvote'].iloc[i] / df['Votes'].iloc[i] if df['Votes'].iloc[i] > 0 else 0.5
+                    actual_b = df['Upvote'].iloc[j] / df['Votes'].iloc[j] if df['Votes'].iloc[j] > 0 else 0.5
+                    df.at[i, 'Elo'] += round(32 * (actual_a - expected_a))
+                    df.at[j, 'Elo'] += round(32 * (actual_b - expected_b))
                 except Exception as e:
                     print(f"Error in ELO calculation for rows {i} and {j}: {str(e)}")
                     continue
-    df['elo'] = round(df['elo'])
-    df['elo_diff'] = df['elo']
+
+        # add link to vote dataset
+        orig_name = df['OrigName'].iloc[i]
+        if ('/' in orig_name):
+            # win rate dataset
+            df.at[i, 'Win Rate'] = f'<a target="_blank" href="https://huggingface.co/datasets/{DB_DATASET_ID}/viewer/summary/rejections?f[rejected][value]=%27{orig_name}%27">' + df['Win Rate'].iloc[i] + '</a>'
+            # df['Win Rate'].iloc[i] = f'<a target="_blank" href="https://huggingface.co/datasets/{DB_DATASET_ID}/viewer/summary/rejections?f[rejected][value]=%27{orig_name}%27">' + df['Win Rate'].iloc[i] + '</a>'
+            # df.loc[i,"Win Rate"] = f'<a target="_blank" href="https://huggingface.co/datasets/{DB_DATASET_ID}/viewer/summary/rejections?f[rejected][value]=%27{orig_name}%27">' + df['Win Rate'].iloc[i] + '</a>'
+    df['Elo'] = round(df['Elo'])
+    df['Elo Diff'] = df['Elo']
 
     if (
         reveal_prelim == False
@@ -57,7 +68,7 @@ def get_leaderboard(reveal_prelim = False):
     try:
         if (reveal_prelim == False):
             for i in range(len(df)):
-                elo_diff = (df['elo'].iloc[i] - leaderboard_df['elo'].iloc[i])
+                elo_diff = (df['Elo'].iloc[i] - leaderboard_df['Elo'].iloc[i])
                 if (elo_diff == 0):
                     continue
                 if (elo_diff > 0):
@@ -65,17 +76,24 @@ def get_leaderboard(reveal_prelim = False):
                 else:
                     plus = '<em style="color: red; font-family: monospace">'
 
-                df.at[i, 'elo_diff'] = str(df['elo'].iloc[i]) + plus + str(elo_diff) +'</em>'
+                df.at[i, 'Elo Diff'] = str(df['Elo'].iloc[i]) + plus + str(elo_diff) +'</em>'
     except:
         # FIXME: crashes when a TTS from premilinary results passes the vote threshold
         pass
 
-    ## ELO score
-    df = df.sort_values(by='elo', ascending=False)
+    # Sort by ELO score and then by Vote count
+    df = df.sort_values(by=['Elo', 'Votes'], ascending=[False, False])
+
+    # last in model series from models.py
+    last_series = []
     # medals
-    def assign_medal(rank, assign):
+    def assign_medal(rank, model_series, last_series, assign):
         rank = str(rank + 1)
-        if assign:
+        if (
+            assign
+            # and last_series != ''
+            # and model_series != last_series
+        ):
             if rank == '1':
                 rank += '🥇'
             elif rank == '2':
@@ -83,11 +101,30 @@ def get_leaderboard(reveal_prelim = False):
             elif rank == '3':
                 rank += '🥉'
 
+        if (model_series != last_series):
+            last_series = model_series
         return '#'+ rank
 
-    df['order'] = [assign_medal(i, not reveal_prelim and len(df) > 2) for i in range(len(df))]
+    try:
+        df['Order'] = [
+            assign_medal(
+                i,
+                HF_SPACES[df['OrigName'].iloc[i]]['series'],
+                last_series,
+                not reveal_prelim and len(df) > 2
+            ) for i in range(len(df))
+        ]
+    except:
+        df['Order'] = [assign_medal(
+            i,
+            '',
+            '',
+            not reveal_prelim and len(df) > 2) for i in range(len(df))
+        ]
+        pass
+
     # fetch top_five
-    for orig_name in df['orig_name']:
+    for orig_name in df['OrigName']:
         if (
             reveal_prelim
             and len(top_five) < 5
@@ -95,6 +132,6 @@ def get_leaderboard(reveal_prelim = False):
         ):
             top_five.append(orig_name)
 
-    df['elo'] = df['elo_diff']
-    df = df[['order', 'name', 'win_rate', 'elo', 'votes']]
+    df['Elo'] = df['Elo Diff']
+    df = df[['Order', 'Name', 'Win Rate', 'Elo', 'Votes']]
     return df

app/ui_leaderboard.py

@@ -10,7 +10,7 @@ with gr.Blocks() as leaderboard:
         min_width=0,
         wrap=False,
         column_widths=[30, 200, 50, 50, 50],
-        datatype=["str", "html", "number", "html", "html"]
+        datatype=["str", "html", "html", "html", "html"],
     )
     reloadbtn = gr.Button("Refresh")
     with gr.Row():