Update app.py

app.py

@@ -91,12 +91,9 @@ def _format_value_minimal(v) -> str:
     if isinstance(v, (float, np.floating)):
         if abs(v - round(v)) < 1e-9:
             return str(int(round(v)))
-        s = f"{float(v):.6f}".rstrip("0").rstrip(".")
-        return s
-    try:
-        return str(v)
-    except Exception:
-        return ""
+    s = f"{float(v):.6f}".rstrip("0").rstrip(".")
+    return s
+
 
 
 def _prepare_dataframe(json_path: str) -> pd.DataFrame:
@@ -231,6 +228,89 @@ def build_view(json_path: str, name_filter: str = "", param_bins: list[str] | No
     return table_value
 
 
+def build_view_and_tasks(
+    json_path: str,
+    name_filter: str = "",
+    param_bins: list[str] | None = None,
+    excluded_tasks: list[str] | None = None,
+):
+    """Return the table and an update object for the exclude-tasks control.
+
+    - The available task choices are derived from the columns of the prepared dataframe
+      by excluding meta columns and helper columns.
+    - The table excludes the selected tasks and recomputes avg_score from only the included tasks.
+    """
+    df = _prepare_dataframe(json_path)
+
+    # Determine all task-like columns (before exclusion)
+    meta_cols_base = [c for c in ["Model", "Provider", "Parameters", "—", "avg_score"] if c in df.columns]
+    tasks_all = [c for c in df.columns if c not in meta_cols_base]
+
+    excluded_set = set(excluded_tasks or [])
+    # Keep only tasks that actually exist
+    excluded_valid = [t for t in excluded_set if t in tasks_all]
+    included_tasks = [c for c in tasks_all if c not in excluded_set]
+
+    # Drop rows that are missing values for required tasks (only those that are included)
+    required_cols = [c for c in ["src_clf", "sum_rag", "sum_rag_v2"] if c in included_tasks]
+    if required_cols:
+        df = df.dropna(subset=required_cols, axis=0)
+
+    # Apply filters
+    df = _apply_filters(df, name_filter=name_filter, param_bins=param_bins)
+
+    # Remove excluded task columns from view
+    if excluded_valid:
+        df = df.drop(columns=[c for c in excluded_valid if c in df.columns], errors="ignore")
+
+    # Recompute avg_score from only included tasks
+    # Determine tasks present in df after exclusion
+    meta_cols_after = [c for c in ["Model", "Provider", "Parameters", "—", "avg_score"] if c in df.columns]
+    current_metric_cols = [c for c in df.columns if c not in meta_cols_after]
+
+    # Drop existing avg_score before recomputation
+    if "avg_score" in df.columns:
+        df = df.drop(columns=["avg_score"])  # will be re-added below
+
+    if current_metric_cols:
+        numeric_df = pd.DataFrame({c: pd.to_numeric(df[c], errors="coerce") for c in current_metric_cols})
+        df["avg_score"] = numeric_df.mean(axis=1, skipna=True).round(2)
+    else:
+        # No metrics left; fill avg_score with NaN to keep schema consistent
+        df["avg_score"] = np.nan
+
+    # Sort and reorder columns similar to _prepare_dataframe
+    if "avg_score" in df.columns:
+        df = df.sort_values(by="avg_score", ascending=False, na_position="last")
+
+    preferred_order = [c for c in ["Model", "Provider", "Parameters"] if c in df.columns]
+    remaining_cols = [c for c in df.columns if c not in preferred_order]
+    if "avg_score" in remaining_cols:
+        remaining_cols = ["avg_score"] + [c for c in remaining_cols if c != "avg_score"]
+    if preferred_order:
+        df = df[preferred_order + remaining_cols]
+
+    # Ensure separator column exists right after Parameters
+    if "Parameters" in df.columns and "—" not in df.columns:
+        insert_at = df.columns.get_loc("Parameters") + 1
+        df.insert(insert_at, "—", "")
+
+    # Style for display
+    if isinstance(df, pd.DataFrame) and not df.empty:
+        styler = df.style
+        if "Parameters" in df.columns:
+            styler = styler.apply(_style_parameters, subset=["Parameters"])  # type: ignore
+        styler = styler.format(_format_value_minimal)
+        table_value: object = styler
+    else:
+        table_value = pd.DataFrame()
+
+    # Update object for the exclude tasks checkbox
+    tasks_update = gr.update(choices=tasks_all, value=excluded_valid)
+
+    return table_value, tasks_update
+
+
 def ui() -> gr.Blocks:
     with gr.Blocks(title="Model Leaderboard") as demo:
         gr.Markdown("""
@@ -250,26 +330,38 @@ def ui() -> gr.Blocks:
                 value=[],
                 info="Select one or more bins"
             )
+        # Task exclusion control (choices are populated dynamically)
+        excluded_tasks_in = gr.CheckboxGroup(
+            label="Exclude tasks",
+            choices=[],
+            value=[],
+            info="Select tasks to hide; all are shown by default",
+        )
 
         # Non-interactive so Pandas Styler is respected; header sorting remains available
         leaderboard_out = gr.Dataframe(label="Leaderboard", interactive=False)
 
         demo.load(
-            fn=build_view,
-            inputs=[json_path_state, name_filter_in, param_bins_in],
-            outputs=[leaderboard_out],
+            fn=build_view_and_tasks,
+            inputs=[json_path_state, name_filter_in, param_bins_in, excluded_tasks_in],
+            outputs=[leaderboard_out, excluded_tasks_in],
         )
 
         # Recompute table on filter changes
         name_filter_in.change(
-            fn=build_view,
-            inputs=[json_path_state, name_filter_in, param_bins_in],
-            outputs=[leaderboard_out],
+            fn=build_view_and_tasks,
+            inputs=[json_path_state, name_filter_in, param_bins_in, excluded_tasks_in],
+            outputs=[leaderboard_out, excluded_tasks_in],
         )
         param_bins_in.change(
-            fn=build_view,
-            inputs=[json_path_state, name_filter_in, param_bins_in],
-            outputs=[leaderboard_out],
+            fn=build_view_and_tasks,
+            inputs=[json_path_state, name_filter_in, param_bins_in, excluded_tasks_in],
+            outputs=[leaderboard_out, excluded_tasks_in],
+        )
+        excluded_tasks_in.change(
+            fn=build_view_and_tasks,
+            inputs=[json_path_state, name_filter_in, param_bins_in, excluded_tasks_in],
+            outputs=[leaderboard_out, excluded_tasks_in],
         )
 
         gr.Markdown("""