feat: add control buttons for metric display in each category

- Add "Select All", "Clear All", and "Show AVG Only" buttons for each category
- Modify initial display to show only AVG metrics
- Update checkbox group to handle new button actions

This change improves the user experience by:
- Making it easier to switch between different metric views
- Showing simpler initial view with only AVG metrics
- Providing quick access to commonly needed display settings

app.py

@@ -291,30 +291,67 @@ with gr.Blocks() as demo_leaderboard:
             else:
                 label = task_type.value
             with gr.Accordion(label, open=True, elem_classes="accordion"):
+                with gr.Row():
+                    gr.Button("全選択", size="sm").click(
+                        fn=lambda t=task_type.name: toggle_category_columns(t, "all", shown_columns_dict),
+                        outputs=shown_columns_dict[task_type.name]
+                    )
+                    gr.Button("全解除", size="sm").click(
+                        fn=lambda t=task_type.name: toggle_category_columns(t, "none", shown_columns_dict),
+                        outputs=shown_columns_dict[task_type.name]
+                    )
+                    gr.Button("AVGのみ", size="sm").click(
+                        fn=lambda t=task_type.name: toggle_category_columns(t, "avg_only", shown_columns_dict),
+                        outputs=shown_columns_dict[task_type.name]
+                    )
                 with gr.Row(height=110):
                     shown_column = gr.CheckboxGroup(
                         show_label=False,
+                        # choicesは全ての選択肢を含む（既存と同じ）
                         choices=[
                             c.name
                             for c in fields(AutoEvalColumn)
                             if not c.hidden and not c.never_hidden and not c.dummy and c.task_type == task_type
-                            # and not c.average
-                            # or (task_type == TaskType.AVG and c.average)
                         ],
+                        # valueは初期表示としてAVGのみを表示
                         value=[
                             c.name
                             for c in fields(AutoEvalColumn)
-                            if c.displayed_by_default
-                            and not c.hidden
-                            and not c.never_hidden
+                            if not c.hidden and not c.never_hidden 
                             and c.task_type == task_type
-                            # and not c.average
-                            # or (task_type == TaskType.AVG and c.average)
+                            and (
+                                (task_type == TaskType.AVG) or  # AVGカテゴリの場合は全て
+                                (task_type != TaskType.AVG and c.average)  # その他のカテゴリはAVGのみ
+                            )
                         ],
                         elem_id="column-select",
                         container=False,
                     )
                     shown_columns_dict[task_type.name] = shown_column
+                # with gr.Row(height=110):
+                #     shown_column = gr.CheckboxGroup(
+                #         show_label=False,
+                #         choices=[
+                #             c.name
+                #             for c in fields(AutoEvalColumn)
+                #             if not c.hidden and not c.never_hidden and not c.dummy and c.task_type == task_type
+                #             # and not c.average
+                #             # or (task_type == TaskType.AVG and c.average)
+                #         ],
+                #         value=[
+                #             c.name
+                #             for c in fields(AutoEvalColumn)
+                #             if c.displayed_by_default
+                #             and not c.hidden
+                #             and not c.never_hidden
+                #             and c.task_type == task_type
+                #             # and not c.average
+                #             # or (task_type == TaskType.AVG and c.average)
+                #         ],
+                #         elem_id="column-select",
+                #         container=False,
+                #     )
+                #     shown_columns_dict[task_type.name] = shown_column
     with gr.Row():
         filter_columns_type = gr.CheckboxGroup(
             label="Model types",
@@ -412,6 +449,27 @@ with gr.Blocks() as demo_leaderboard:
     # Check query parameter once at startup and update search bar + hidden component
     demo_leaderboard.load(fn=load_query, outputs=[search_bar, hidden_search_bar])
 
+def toggle_category_columns(task_type: str, action: str, shown_columns_dict: dict):
+    """カテゴリーごとのチェックボックスを制御する関数"""
+    columns = [
+        c.name for c in fields(AutoEvalColumn)
+        if not c.hidden and not c.never_hidden and not c.dummy 
+        and c.task_type.name == task_type
+    ]
+    
+    if action == "all":
+        return columns
+    elif action == "none":
+        return []
+    elif action == "avg_only":
+        if task_type == TaskType.AVG.name:
+            return columns  # 全体AVGの場合
+        return [
+            c.name for c in fields(AutoEvalColumn)
+            if not c.hidden and not c.never_hidden and c.average 
+            and c.task_type.name == task_type  # カテゴリー別AVGの場合
+        ]
+
 
 # Submission demo