Unselect task datasets will update average and npm

app.py

@@ -28,7 +28,8 @@ from src.display.utils import (
     ModelType,
     fields,
     WeightType,
-    Precision
+    Precision,
+    Tasks
 )
 from src.envs import (
     API,
@@ -126,6 +127,7 @@ def update_table(
 ):
     filtered_df = filter_models(df=hidden_df, type_query=type_query, size_query=size_query, precision_query=precision_query, hide_models=hide_models)
     filtered_df = filter_queries(query, filtered_df)
+    filtered_df = update_leaderboard_avg_scores(filtered_df, columns)
     df = select_columns(filtered_df, columns)
     return df
 
@@ -200,6 +202,21 @@ def filter_models(
 
     return filtered_df
 
+def update_leaderboard_avg_scores(df, columns):
+    new_df = df.copy()
+
+    #update average with tasks in shown columns
+    task_columns = []
+    task_baseline = []
+    for task in Tasks:
+        column_name = getattr(AutoEvalColumn, task.name).name
+        if column_name in columns:
+            task_columns.append(column_name)
+            task_baseline.append(task.value.baseline)
+    new_df[AutoEvalColumn.average.name] = new_df[task_columns].mean(axis=1).apply(lambda x: round(x, 2))
+    new_df[AutoEvalColumn.npm.name] = (((new_df[task_columns] - task_baseline) / [100.0 - t for t in task_baseline]).mean(axis=1) * 100).apply(lambda x: round(x, 2))
+    return new_df
+
 leaderboard_df = filter_models(
     df=leaderboard_df, 
     type_query=[t.to_str(" : ") for t in ModelType],