Spaces:
Sleeping
Sleeping
| #!/usr/bin/env python3 | |
| """Results analyser - loads trial results and generates paper tables. | |
| Usage: | |
| python run_analysis.py # uses data/results/trial_results.json | |
| python run_analysis.py --session path/to.csv # use a session CSV instead | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import json | |
| import logging | |
| from pathlib import Path | |
| logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s") | |
| logger = logging.getLogger(__name__) | |
| def main() -> None: | |
| parser = argparse.ArgumentParser(description="OmniNPC Results Analyser") | |
| parser.add_argument("--session", type=str, default=None, help="Path to session CSV") | |
| parser.add_argument("--baseline", type=str, default=None, help="Path to baseline CSV") | |
| args = parser.parse_args() | |
| from evaluation.results_analyser import ( | |
| analyse_session, generate_paper_table, save_paper_results, | |
| ) | |
| # Prefer trial_results.json (from run_evaluation.py) — these are real benchmark numbers | |
| trial_json = Path("data/results/trial_results.json") | |
| results = None | |
| if trial_json.exists(): | |
| with open(trial_json, encoding="utf-8") as f: | |
| results = json.load(f) | |
| logger.info("Loaded trial results from %s (timestamp: %s)", | |
| trial_json, results.get("timestamp", "unknown")) | |
| else: | |
| # Fall back to session CSV analysis | |
| csv_path = args.session | |
| if csv_path is None: | |
| log_dir = Path("data/logs") | |
| csv_files = sorted(log_dir.glob("session_*.csv")) | |
| if csv_files: | |
| csv_path = str(csv_files[-1]) | |
| logger.info("Using most recent session: %s", csv_path) | |
| else: | |
| logger.error( | |
| "No trial_results.json found and no session CSV in data/logs/. " | |
| "Run: python run_evaluation.py --trials all --n 50" | |
| ) | |
| return | |
| results = analyse_session(csv_path) | |
| baseline = None | |
| if args.baseline: | |
| baseline = analyse_session(args.baseline) | |
| table = generate_paper_table(results, baseline) | |
| print("\n" + table) | |
| save_paper_results(table, "data/results/PAPER_RESULTS.txt") | |
| logger.info("Paper results written to data/results/PAPER_RESULTS.txt") | |
| if __name__ == "__main__": | |
| main() | |