Update src/populate.py

src/populate.py

@@ -14,7 +14,28 @@ def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchm
 
     df = pd.DataFrame.from_records(all_data_json)
     # df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
-    df = df.sort_values(by=[AutoEvalColumn.task5.name], ascending=True)
+    # df = df.sort_values(by=[AutoEvalColumn.task5.name], ascending=True)
+
+    df[AutoEvalColumn.task3.name] = pd.Series(
+        np.stack(
+            np.array(df[AutoEvalColumn.task3.name].values)
+        ).squeeze()
+    )
+    df[AutoEvalColumn.task4.name] = pd.Series(
+        np.stack(
+            np.array(df[AutoEvalColumn.task4.name].values)
+        ).squeeze()
+    )
+
+    
+    mos_rank = df[AutoEvalColumn.task3.name].rank(method="min", numeric_only=True, ascending=False)
+    bitrate_rank = df[AutoEvalColumn.task4.name].rank(method="min", numeric_only=True, ascending=True)
+    df["Ranking"] = pd.Series((mos_rank + bitrate_rank)/2)
+    df = df.sort_values(by=["Ranking", AutoEvalColumn.task2.name], ascending=True)
+    df["Rank"] = df.groupby("Precision").cumcount() + 1
+    df.pop("Ranking")
+
+    
     df = df[cols].round(decimals=2)
 
     # filter out if any of the benchmarks have not been produced