Update app.py

app.py

@@ -53,6 +53,11 @@ except Exception:
     restart_space()
 
 LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
+print(f"Initial leaderboard_df shape: {LEADERBOARD_DF.shape}")
+print("Initial leaderboard_df columns:", LEADERBOARD_DF.columns.tolist())
+print("Initial leaderboard_df sample data:")
+print(LEADERBOARD_DF.head())
+
 original_df = LEADERBOARD_DF
 leaderboard_df = original_df.copy()
 (
@@ -76,15 +81,25 @@ def update_table(
     show_flagged: bool,
     query: str,
 ):
-    filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, add_special_tokens_query, num_few_shots_query, show_deleted, show_merges, show_flagged)
+    filtered_df = filter_models(
+        hidden_df, type_query, size_query, precision_query, 
+        add_special_tokens_query, num_few_shots_query, 
+        show_deleted, show_merges, show_flagged
+    )
     filtered_df = filter_queries(query, filtered_df)
+    
     print(f"Filter applied: query={query}, columns={columns}, type_query={type_query}, precision_query={precision_query}")
+    print(f"Filtered DataFrame shape: {filtered_df.shape}")
     print(filtered_df.head())  # フィルタ後のデータを確認
-
+    
     df = select_columns(filtered_df, columns)
+    print(f"DataFrame after selecting columns: {df.shape}")
+    print(df.head())  # 選択後のデータを確認
+    
     return df
 
 
+
 def load_query(request: gr.Request):  # triggered only once at startup => read query parameter if it exists
     query = request.query_params.get("query") or ""
     return query, query # return one for the "search_bar", one for a hidden component that triggers a reload only if value has changed
@@ -127,33 +142,64 @@ 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
+    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:
-    # Show all models
+    # 初期状態
+    print(f"Initial DataFrame shape: {df.shape}")
+    
+    # Show deleted models フィルタ
     if show_deleted:
         filtered_df = df
-    else:  # Show only still on the hub models
+        print("Show deleted models: ON")
+    else:
         filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
-
-    #if not show_merges:
-    #    filtered_df = filtered_df[filtered_df[AutoEvalColumn.merged.name] == False]
-
-    #if not show_flagged:
-    #    filtered_df = filtered_df[filtered_df[AutoEvalColumn.flagged.name] == False]
-
+        print(f"After filtering deleted models: {filtered_df.shape}")
+    
+    # Show merges フィルタ（コメントアウトされている場合はスキップ）
+    # if not show_merges:
+    #     filtered_df = filtered_df[filtered_df[AutoEvalColumn.merged.name] == False]
+    #     print(f"After filtering merged models: {filtered_df.shape}")
+    
+    # Show flagged フィルタ（コメントアウトされている場合はスキップ）
+    # if not show_flagged:
+    #     filtered_df = filtered_df[filtered_df[AutoEvalColumn.flagged.name] == False]
+    #     print(f"After filtering flagged models: {filtered_df.shape}")
+    
+    # Model type フィルタ
     type_emoji = [t[0] for t in type_query]
-    filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
-    filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
-    filtered_df = filtered_df.loc[df[AutoEvalColumn.add_special_tokens.name].isin(add_special_tokens_query)]
-    filtered_df = filtered_df.loc[df[AutoEvalColumn.num_few_shots.name].isin(num_few_shots_query)]
+    filtered_df = filtered_df[filtered_df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
+    print(f"After filtering by model type: {filtered_df.shape}")
     
-
+    # Precision フィルタ
+    filtered_df = filtered_df[filtered_df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]
+    print(f"After filtering by precision: {filtered_df.shape}")
+    
+    # Add Special Tokens フィルタ
+    filtered_df = filtered_df[filtered_df[AutoEvalColumn.add_special_tokens.name].isin(add_special_tokens_query)]
+    print(f"After filtering by add_special_tokens: {filtered_df.shape}")
+    
+    # Num Few Shots フィルタ
+    filtered_df = filtered_df[filtered_df[AutoEvalColumn.num_few_shots.name].isin(num_few_shots_query)]
+    print(f"After filtering by num_few_shots: {filtered_df.shape}")
+    
+    # Model size フィルタ
     numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
-    params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
+    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]
+    filtered_df = filtered_df[mask]
+    print(f"After filtering by model size: {filtered_df.shape}")
+    
     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)
 
 demo = gr.Blocks(css=custom_css)