Update app.py

app.py

@@ -4,7 +4,6 @@ import gradio as gr
 import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import snapshot_download
-from decimal import Decimal
 
 from src.about import (
     CITATION_BUTTON_LABEL,
@@ -54,13 +53,7 @@ except Exception:
     restart_space()
 
 LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
-print(LEADERBOARD_DF.head())
 original_df = LEADERBOARD_DF
-print("Initial LEADERBOARD_DF:")
-print(LEADERBOARD_DF.head())
-print(f"LEADERBOARD_DF shape: {LEADERBOARD_DF.shape}")
-print("LEADERBOARD_DF columns:")
-print(LEADERBOARD_DF.columns.tolist())
 leaderboard_df = original_df.copy()
 (
     finished_eval_queue_df,
@@ -83,10 +76,10 @@ 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_queries(query, filtered_df)
+    print(f"Filter applied: query={query}, columns={columns}, type_query={type_query}, precision_query={precision_query}")
+    print(filtered_df.head())  # フィルタ後のデータを確認
 
     df = select_columns(filtered_df, columns)
     return df
@@ -136,75 +129,40 @@ 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}")
-    print(f"Initial df shape: {df.shape}")
-    print("Initial df columns:")
-    print(df.columns.tolist())
-    
-    filtered_df = df.copy()  # Create a copy to avoid modifying the original dataframe
+    # Show all models
+    if show_deleted:
+        filtered_df = df
+    else:  # Show only still on the hub models
+        filtered_df = df[df[AutoEvalColumn.still_on_hub.name] == True]
 
-    # Show all models (comment out for debugging)
-    # if not show_deleted:
-    #     filtered_df = filtered_df[filtered_df[AutoEvalColumn.still_on_hub.name] == True]
+    #if not show_merges:
+    #    filtered_df = filtered_df[filtered_df[AutoEvalColumn.merged.name] == False]
 
-    print(f"After deletion filter: {filtered_df.shape}")
+    #if not show_flagged:
+    #    filtered_df = filtered_df[filtered_df[AutoEvalColumn.flagged.name] == False]
 
-    # Type filter
     type_emoji = [t[0] for t in type_query]
-    if 'T' in filtered_df.columns:
-        filtered_df = filtered_df[filtered_df['T'].isin(type_emoji + ['?'])]  # Include '?' as a valid type
-    elif 'Type_Symbol' in filtered_df.columns:
-        filtered_df = filtered_df[filtered_df['Type_Symbol'].isin(type_emoji + ['?'])]
-    else:
-        print("Warning: Neither 'T' nor 'Type_Symbol' column found in the dataframe")
+    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)]
+    print("Filtered DataFrame shape:", filtered_df.shape)
+    print("Filtered DataFrame columns:", filtered_df.columns.tolist())
+    print("Filtered DataFrame sample:", filtered_df.head().to_dict('records'))
     
-    print(f"After type filter: {filtered_df.shape}")
-
-    # Precision filter
-    precision_query = precision_query + ['Unknown', '?']
-    filtered_df = filtered_df[filtered_df[AutoEvalColumn.precision.name].isin(precision_query)]
-    print(f"After precision filter: {filtered_df.shape}")
-
-    # Add Special Tokens filter
-    add_special_tokens_query = 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}")
-
-    # Num Few Shots filter
-    num_few_shots_query = 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}")
 
-    # Size filter
     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: pd.isna(x) or any(numeric_interval.contains(x)))
-    filtered_df = filtered_df[mask]
-    print(f"After size filter: {filtered_df.shape}")
-
-    if filtered_df.empty:
-        print("Warning: Filtered dataframe is empty!")
-        return pd.DataFrame(columns=df.columns)  # Return an empty dataframe with the same columns
-
-    print("Filtered dataframe head:")
-    print(filtered_df.head())
-    print("Column names:")
-    print(filtered_df.columns.tolist())
-    print("Column data types:")
-    print(filtered_df.dtypes)
-    print("Final filtered dataframe sample:")
-    print(filtered_df.head().to_dict('records'))
-
-    print("Filtered DataFrame sample:")
-    print(filtered_df.head().to_dict('records'))
-
-    filtered_df = filtered_df.astype(str)
+    params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
+    mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
+    filtered_df = filtered_df.loc[mask]
     return filtered_df
 
-def convert_decimal_to_str(item):
-    return {k: str(v) if isinstance(v, Decimal) else v for k, v in item.items()}
+filtered_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)
+display_columns = [c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value
+display_data = filtered_df[display_columns].to_dict('records')
 
-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)
+print("Display columns:", display_columns)
+print("Display data sample:", display_data[:1])
 
 demo = gr.Blocks(css=custom_css)
 with demo:
@@ -285,48 +243,15 @@ with demo:
                         elem_id="filter-columns-num-few-shots",
                     )
 
-            # leaderboard_table = gr.components.Dataframe(
-            #     value=leaderboard_df[
-            #         [c.name for c in fields(AutoEvalColumn) if c.never_hidden]
-            #         + shown_columns.value
-            #         # + [AutoEvalColumn.dummy.name]
-            #     ],
-            #     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,
-            #     #column_widths=["2%", "33%"] 
-            # )
-            filtered_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)
-            print("After filter_models:")
-            print(f"filtered_df shape: {filtered_df.shape}")
-            print("filtered_df columns:")
-            print(filtered_df.columns.tolist())
-
-            initial_data = [convert_decimal_to_str(item) for item in filtered_df.to_dict('records')]
-            headers = filtered_df.columns.tolist()
-            print("Filtered DataFrame contents:")
-            print(filtered_df.head().to_dict('records'))
-            print("Filtered DataFrame columns:")
-            print(filtered_df.columns.tolist())
-            filtered_df_without_T = filtered_df.drop('T', axis=1)
             leaderboard_table = gr.components.Dataframe(
-                value=filtered_df_without_T.to_dict('records'),
-                headers=filtered_df_without_T.columns.tolist(),
-                datatype={col: "str" for col in filtered_df_without_T.columns},
-                row_count=(len(filtered_df_without_T), "dynamic"),
-                col_count=(len(filtered_df_without_T.columns), "fixed"),
-                wrap=True,
+                value=display_data,
+                headers=display_columns,
+                datatype={col: str(TYPES.get(col, "str")) for col in display_columns},
                 elem_id="leaderboard-table",
                 interactive=False,
                 visible=True,
             )
-            print("Leaderboard table initial value:")
-            print(initial_data[:5] if initial_data else "Empty")
-            print("Headers:")
-            print(headers)
-            print("After Dataframe initialization")
+            print(leaderboard_df.head())  # リーダーボードテーブルに渡される前のデータを確認
 
             # Dummy leaderboard for handling the case when the user uses backspace key
             hidden_leaderboard_table_for_search = gr.components.Dataframe(