Hide extra columns

fev_pages/fev_bench.py

@@ -58,6 +58,10 @@ MODEL_TYPES_HELP = (
     'Click "See details" below the leaderboard for a description of each model type.'
 )
 NON_COMMERCIAL_HELP = "Include models whose license does not permit commercial use."
+IMPUTATION_COLUMNS_HELP = (
+    "Percentage of tasks where the model's training corpus overlaps with the task (scores replaced "
+    f"with {LEAKAGE_IMPUTATION_MODEL}) or the model failed (scores replaced with {BASELINE_MODEL})."
+)
 
 # Group type options
 GROUP_TYPES = [
@@ -202,6 +206,12 @@ with st.sidebar:
         )
         st.caption(f"{num_models} models")
 
+    st.divider()
+
+    show_imputation_columns = st.toggle(
+        "Show leakage & failure columns", value=False, help=IMPUTATION_COLUMNS_HELP
+    )
+
 cols = st.columns(spec=[0.025, 0.95, 0.025])
 
 with cols[1] as main_container:
@@ -231,6 +241,12 @@ with cols[1] as main_container:
         df_styled,
         width="stretch",
         hide_index=True,
+        column_order=[
+            col
+            for col in df_styled.data.columns
+            if show_imputation_columns
+            or col not in ("training_corpus_overlap", "num_failures")
+        ],
         column_config={
             "model_name": st.column_config.LinkColumn(
                 label="Model", display_text=r"#(.*)$"

src/utils.py

@@ -170,10 +170,10 @@ def format_leaderboard(df: pd.DataFrame):
             "win_rate",
             "skill_score",
             "median_inference_time_s_per100",
-            "training_corpus_overlap",
-            "num_failures",
             "org",
             "tags",
+            "training_corpus_overlap",
+            "num_failures",
         ]
     ]
     return df.style.apply(