Update app.py

app.py

@@ -143,43 +143,37 @@ def filter_queries(query: str, filtered_df: pd.DataFrame):
 def filter_models(
     df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, add_special_tokens_query: list, num_few_shots_query: list, show_deleted: bool, show_merges: bool, show_flagged: bool
 ) -> pd.DataFrame:
+    print(f"filter_models called with: type_query={type_query}, size_query={size_query}, precision_query={precision_query}, add_special_tokens_query={add_special_tokens_query}, num_few_shots_query={num_few_shots_query}")
     print(f"Initial df shape: {df.shape}")
-    print(f"Initial df columns: {df.columns}")
     print(f"Initial df content:\n{df}")
-    
-    filtered_df = df
 
-    # モデルタイプでフィルタリング
-    type_emoji = [t.split()[0] for t in type_query]
-    filtered_df = filtered_df[filtered_df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji + ["?"])]
+    filtered_df = df
+    
+    # type_emoji = [t[0] for t in type_query]
+    # filtered_df = filtered_df[filtered_df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
     print(f"After type filter: {filtered_df.shape}")
+    print(f"After type filter content:\n{filtered_df}")
 
-    # Precisionでフィルタリング
-    filtered_df = filtered_df[filtered_df[AutoEvalColumn.precision.name].isin(precision_query + ["None", "Unknown", "?"])]
+    # Precision filterをコメントアウト
+    # filtered_df = filtered_df[filtered_df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
     print(f"After precision filter: {filtered_df.shape}")
+    print(f"After precision filter content:\n{filtered_df}")
 
-    # Add Special Tokensでフィルタリング
-    filtered_df = filtered_df[filtered_df[AutoEvalColumn.add_special_tokens.name].isin(add_special_tokens_query + ["Unknown", "?"])]
+    # filtered_df = filtered_df[filtered_df[AutoEvalColumn.add_special_tokens.name].isin(add_special_tokens_query)]
     print(f"After add_special_tokens filter: {filtered_df.shape}")
+    print(f"After add_special_tokens filter content:\n{filtered_df}")
 
-    # Num Few Shotsでフィルタリング
-    filtered_df = filtered_df[filtered_df[AutoEvalColumn.num_few_shots.name].isin(num_few_shots_query + ["Unknown", "?"])]
+    # filtered_df = filtered_df[filtered_df[AutoEvalColumn.num_few_shots.name].isin(num_few_shots_query)]
     print(f"After num_few_shots filter: {filtered_df.shape}")
+    print(f"After num_few_shots filter content:\n{filtered_df}")
 
-    # モデルサイズでフィルタリング
-    if "Unknown" in size_query:
-        size_mask = filtered_df[AutoEvalColumn.params.name].isna() | (filtered_df[AutoEvalColumn.params.name].astype(str).isin(["Unknown", "?"]))
-    else:
-        numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query if s != "Unknown"]))
-        params_column = pd.to_numeric(filtered_df[AutoEvalColumn.params.name], errors="coerce")
-        size_mask = params_column.apply(lambda x: any(numeric_interval.contains(x)) if pd.notnull(x) else False)
-    
-    filtered_df = filtered_df[size_mask]
+    # numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
+    # params_column = pd.to_numeric(filtered_df[AutoEvalColumn.params.name], errors="coerce")
+    # mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
+    # filtered_df = filtered_df.loc[mask]
     print(f"After size filter: {filtered_df.shape}")
+    print(f"After size filter content:\n{filtered_df}")
 
-    print("Filtered dataframe head:")
-    print(filtered_df.head())
-    print(f"Filtered df columns: {filtered_df.columns}")
     return filtered_df
 
 leaderboard_df = filter_models(leaderboard_df, [t.to_str(" : ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], [i.value.name for i in AddSpecialTokens], [i.value.name for i in NumFewShots], False, False, False)
@@ -263,22 +257,19 @@ with demo:
                         elem_id="filter-columns-num-few-shots",
                     )
 
-            leaderboard_df_filtered = filter_models(leaderboard_df, [t.to_str(" ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], [i.value.name for i in AddSpecialTokens], [i.value.name for i in NumFewShots], False, False, False)
+            leaderboard_df_filtered = filter_models(leaderboard_df, [t.to_str(" : ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], [i.value.name for i in AddSpecialTokens], [i.value.name for i in NumFewShots], False, False, False)
 
-            # 列名の重複を解消
-            leaderboard_df_filtered.columns = pd.io.parsers.base._maybe_dedup_names(leaderboard_df_filtered.columns)
-            
             leaderboard_table = gr.components.Dataframe(
-                value=leaderboard_df_filtered.to_dict('records'),
-                headers=list(leaderboard_df_filtered.columns),
-                datatype={col: "str" for col in leaderboard_df_filtered.columns},
+                value=leaderboard_df_filtered,
+                headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
+                datatype=TYPES,
                 elem_id="leaderboard-table",
                 interactive=False,
                 visible=True,
             )
             print("Leaderboard table initial value:")
-            print(leaderboard_df_filtered)
-            print(f"Leaderboard table shape: {leaderboard_df_filtered.shape}")
+            print(leaderboard_table.value)
+            print(f"Leaderboard table shape: {leaderboard_table.value.shape if isinstance(leaderboard_table.value, pd.DataFrame) else 'Not a DataFrame'}")
             
             # Dummy leaderboard for handling the case when the user uses backspace key
             hidden_leaderboard_table_for_search = gr.components.Dataframe(