Datasets:
Nikame Agent
fix: removed all duplicate candidates across the benchmark using rigorous multi-pass review
696b5d1 | import json | |
| import os | |
| import glob | |
| import random | |
| DATA_DIR = '/mnt/data/projects/GomParam-v1/data' | |
| def fix_duplicates(candidates, correct_idx): | |
| # A generic list of safe distractors across different topics just in case | |
| # we'll try to use distractors from the same file's other candidates first | |
| pass | |
| def process_file(filepath): | |
| with open(filepath, 'r', encoding='utf-8') as f: | |
| data = json.load(f) | |
| # collect all possible candidates in this file to use as a pool | |
| all_cands = set() | |
| for item in data: | |
| if 'candidates' in item: | |
| for c in item['candidates']: | |
| all_cands.add(c) | |
| all_cands = list(all_cands) | |
| modified = False | |
| for item in data: | |
| if 'candidates' in item: | |
| cands = item['candidates'] | |
| correct_idx = item.get('correct') | |
| # If duplicates exist | |
| if len(set(cands)) != len(cands): | |
| new_cands = [] | |
| seen = set() | |
| # Keep the correct answer as is, and keep its position! | |
| # To do this safely: | |
| for i, c in enumerate(cands): | |
| if i == correct_idx: | |
| new_cands.append(c) | |
| seen.add(c) | |
| else: | |
| new_cands.append(None) # placeholder | |
| for i, c in enumerate(cands): | |
| if i != correct_idx: | |
| if c not in seen: | |
| new_cands[i] = c | |
| seen.add(c) | |
| else: | |
| # Need a replacement | |
| while True: | |
| repl = random.choice(all_cands) | |
| if repl not in seen: | |
| new_cands[i] = repl | |
| seen.add(repl) | |
| break | |
| item['candidates'] = new_cands | |
| modified = True | |
| print(f"Fixed {item['id']}: {cands} -> {new_cands}") | |
| if modified: | |
| with open(filepath, 'w', encoding='utf-8') as f: | |
| json.dump(data, f, ensure_ascii=False, indent=2) | |
| def main(): | |
| for file in glob.glob(os.path.join(DATA_DIR, '*.json')): | |
| process_file(file) | |
| if __name__ == '__main__': | |
| main() | |