Spaces:
Runtime error
Runtime error
| import pandas as pd | |
| import sys | |
| from src.display.formatting import has_no_nan_values, make_clickable_model | |
| from src.display.utils import AutoEvalColumn | |
| from src.leaderboard.read_evals import get_raw_eval_results | |
| def get_leaderboard_df(results_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame: | |
| """Creates a dataframe from all the individual experiment results""" | |
| try: | |
| sys.stderr.write("\n=== GET_LEADERBOARD_DF DEBUG ===\n") | |
| sys.stderr.write("Starting leaderboard creation...\n") | |
| sys.stderr.write(f"Looking for results in: {results_path}\n") | |
| sys.stderr.write(f"Expected columns: {cols}\n") | |
| sys.stderr.write(f"Benchmark columns: {benchmark_cols}\n") | |
| sys.stderr.flush() | |
| raw_data = get_raw_eval_results(results_path) | |
| sys.stderr.write(f"\nFound {len(raw_data)} raw results\n") | |
| sys.stderr.flush() | |
| if not raw_data: | |
| sys.stderr.write("No raw data found, creating empty DataFrame\n") | |
| sys.stderr.flush() | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| all_data_json = [] | |
| for i, v in enumerate(raw_data): | |
| try: | |
| sys.stderr.write(f"Processing result {i+1}/{len(raw_data)}: {v.full_model}\n") | |
| sys.stderr.flush() | |
| data_dict = v.to_dict() | |
| # Validate the data_dict has required columns | |
| missing_cols = [col for col in cols if col not in data_dict] | |
| if missing_cols: | |
| sys.stderr.write(f"WARNING: Result for {v.full_model} missing columns: {missing_cols}\n") | |
| # Add missing columns with default values | |
| for col in missing_cols: | |
| if col in benchmark_cols: | |
| data_dict[col] = 0.0 | |
| elif col == AutoEvalColumn.model_type_symbol.name: | |
| data_dict[col] = "?" | |
| else: | |
| data_dict[col] = "" | |
| sys.stderr.flush() | |
| all_data_json.append(data_dict) | |
| sys.stderr.write(f"Successfully processed result {i+1}/{len(raw_data)}: {v.full_model}\n") | |
| sys.stderr.flush() | |
| except Exception as e: | |
| sys.stderr.write(f"Error processing result {i+1}/{len(raw_data)} ({v.full_model}): {e}\n") | |
| import traceback | |
| sys.stderr.write(f"Traceback: {traceback.format_exc()}\n") | |
| sys.stderr.flush() | |
| continue | |
| sys.stderr.write(f"\nConverted to {len(all_data_json)} JSON records\n") | |
| sys.stderr.flush() | |
| if not all_data_json: | |
| sys.stderr.write("No valid JSON records, creating empty DataFrame\n") | |
| sys.stderr.flush() | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| if all_data_json: | |
| sys.stderr.write("Sample record keys: " + str(list(all_data_json[0].keys())) + "\n") | |
| sys.stderr.flush() | |
| try: | |
| df = pd.DataFrame.from_records(all_data_json) | |
| sys.stderr.write("\nCreated DataFrame with columns: " + str(df.columns.tolist()) + "\n") | |
| sys.stderr.write("DataFrame shape: " + str(df.shape) + "\n") | |
| sys.stderr.flush() | |
| except Exception as e: | |
| sys.stderr.write(f"Error creating DataFrame from records: {e}\n") | |
| sys.stderr.flush() | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| try: | |
| # No sorting needed - we only have p-values | |
| sys.stderr.write("\nNo sorting applied - only p-values\n") | |
| sys.stderr.flush() | |
| except Exception as e: | |
| sys.stderr.write(f"\nError with DataFrame: {e}\n") | |
| sys.stderr.write("Available columns: " + str(df.columns.tolist()) + "\n") | |
| sys.stderr.flush() | |
| try: | |
| # Ensure all required columns exist before selecting | |
| for col in cols: | |
| if col not in df.columns: | |
| sys.stderr.write(f"Adding missing column during selection: {col}\n") | |
| if col in benchmark_cols or col == AutoEvalColumn.average.name: | |
| df[col] = 0.0 | |
| else: | |
| df[col] = "" | |
| sys.stderr.flush() | |
| df = df[cols].round(decimals=2) | |
| sys.stderr.write("\nSelected and rounded columns\n") | |
| sys.stderr.flush() | |
| except Exception as e: | |
| sys.stderr.write(f"\nError selecting columns: {e}\n") | |
| sys.stderr.write("Requested columns: " + str(cols) + "\n") | |
| sys.stderr.write("Available columns: " + str(df.columns.tolist()) + "\n") | |
| sys.stderr.flush() | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| # No filtering needed - we only have p-values | |
| sys.stderr.write("\nFinal DataFrame shape (no filtering): " + str(df.shape) + "\n") | |
| sys.stderr.write("Final columns: " + str(df.columns.tolist()) + "\n") | |
| sys.stderr.flush() | |
| # Final validation | |
| if df is None or df.empty: | |
| sys.stderr.write("Final DataFrame is None or empty, returning fallback\n") | |
| sys.stderr.flush() | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| sys.stderr.write(f"=== FINAL RESULT: DataFrame with {len(df)} rows and {len(df.columns)} columns ===\n") | |
| sys.stderr.flush() | |
| return df | |
| except Exception as e: | |
| sys.stderr.write(f"\nCRITICAL ERROR in get_leaderboard_df: {e}\n") | |
| import traceback | |
| sys.stderr.write(f"Traceback: {traceback.format_exc()}\n") | |
| sys.stderr.flush() | |
| # Always return a valid DataFrame, never None | |
| return create_empty_dataframe(cols, benchmark_cols) | |
| def create_empty_dataframe(cols: list, benchmark_cols: list) -> pd.DataFrame: | |
| """Create a valid empty DataFrame with all required columns""" | |
| import sys | |
| sys.stderr.write("Creating empty fallback DataFrame...\n") | |
| sys.stderr.flush() | |
| empty_df = pd.DataFrame(columns=cols) | |
| # Ensure correct column types | |
| for col in cols: | |
| if col in benchmark_cols: | |
| empty_df[col] = pd.Series(dtype=float) | |
| else: | |
| empty_df[col] = pd.Series(dtype=str) | |
| sys.stderr.write(f"Empty DataFrame created with columns: {empty_df.columns.tolist()}\n") | |
| sys.stderr.flush() | |
| return empty_df | |