Download scripts/update_problem_count.py from FrontierCS/Frontier-CS: direct link, hf CLI and curl.
- Browser
- Download file 3.31 kB
-
https://huggingface.co/datasets/FrontierCS/Frontier-CS/resolve/2da8cfa39b07a3ddd5bd46be7448f13d8d91cfb8/scripts/update_problem_count.py
- Command line
-
hf download hf://datasets/FrontierCS/Frontier-CS@2da8cfa39b07a3ddd5bd46be7448f13d8d91cfb8/scripts/update_problem_count.py
-
curl -L -o update_problem_count.py https://huggingface.co/datasets/FrontierCS/Frontier-CS/resolve/2da8cfa39b07a3ddd5bd46be7448f13d8d91cfb8/scripts/update_problem_count.py
3.31 kB
| #!/usr/bin/env python3 | |
| """ | |
| Script to automatically count and update problem statistics in README.md | |
| Run this script to update the problem counts in the README. | |
| """ | |
| import os | |
| from pathlib import Path | |
| def count_research_problems(research_dir: Path) -> int: | |
| """Count research problems by counting evaluator.py files, with special handling for poc_generation.""" | |
| count = 0 | |
| # Special case: poc_generation counts as 4 problems (4 subcategories) | |
| poc_dir = research_dir / 'problems' / 'poc_generation' | |
| if poc_dir.exists(): | |
| count += 4 | |
| # Count all evaluator.py files, excluding those in poc_generation | |
| for evaluator_file in research_dir.rglob('evaluator.py'): | |
| # Skip if it's under poc_generation directory | |
| if 'poc_generation' not in str(evaluator_file): | |
| count += 1 | |
| return count | |
| def count_algorithmic_problems(algorithmic_dir: Path) -> int: | |
| """Count algorithmic problems by counting numbered directories.""" | |
| problems_dir = algorithmic_dir / 'problems' | |
| if not problems_dir.exists(): | |
| return 0 | |
| count = 0 | |
| for item in problems_dir.iterdir(): | |
| if item.is_dir() and item.name.isdigit(): | |
| count += 1 | |
| return count | |
| def update_readme_badge(readme_path: Path, research_count: int, algo_count: int): | |
| """Update the README with current problem counts using badges.""" | |
| if not readme_path.exists(): | |
| print(f"README not found at {readme_path}") | |
| return | |
| content = readme_path.read_text() | |
| # Create the research and algorithmic badges (HTML img tag format) | |
| research_badge = f'<img src="https://img.shields.io/badge/Research_Problems-{research_count}-blue" alt="Research Problems">' | |
| algo_badge = f'<img src="https://img.shields.io/badge/Algorithmic_Problems-{algo_count}-green" alt="Algorithmic Problems">' | |
| # Replace existing badges using regex | |
| import re | |
| # Replace Research Problems badge (HTML img tag format) | |
| content = re.sub( | |
| r'<img src="https://img\.shields\.io/badge/Research_Problems-\d+-blue" alt="Research Problems">', | |
| research_badge, | |
| content | |
| ) | |
| # Replace Algorithmic Problems badge (HTML img tag format) | |
| content = re.sub( | |
| r'<img src="https://img\.shields\.io/badge/Algorithmic_Problems-\d+-green" alt="Algorithmic Problems">', | |
| algo_badge, | |
| content | |
| ) | |
| # Write back | |
| readme_path.write_text(content) | |
| print(f"✅ Updated README with counts: Research={research_count}, Algorithmic={algo_count}") | |
| def main(): | |
| # Get repository root | |
| script_dir = Path(__file__).parent | |
| repo_root = script_dir.parent | |
| # Count problems | |
| research_dir = repo_root / 'research' | |
| algorithmic_dir = repo_root / 'algorithmic' | |
| readme_path = repo_root / 'README.md' | |
| research_count = count_research_problems(research_dir) | |
| algo_count = count_algorithmic_problems(algorithmic_dir) | |
| print(f"📊 Problem Statistics:") | |
| print(f" Research Problems: {research_count}") | |
| print(f" Algorithmic Problems: {algo_count}") | |
| print(f" Total: {research_count + algo_count}") | |
| # Update README | |
| update_readme_badge(readme_path, research_count, algo_count) | |
| if __name__ == '__main__': | |
| main() | |