ml4se

COSC 493: Machine Learning for Software Engineering

Predictive models to identify potentially buggy code.

Repository map

• GithubScraper/ - notebooks and scripts for collecting GitHub Java projects.
• StackOverflowScraper/ - StackOverflow Java question/answer scraper.
• parse_source_code/ - BabelFish/UAST experiments for source parsing.
• repo_history/ - utilities for creating before/after commit file pairs.
• blackbox_python/ - small Python package experiment and example code.

Development notes

Many scripts target older notebook, scraping, or BabelFish workflows. Treat this as a course/research archive unless dependencies are refreshed for a specific experiment. Generated notebook checkpoints, IDE metadata, lock files, and Python bytecode are ignored.

Reproducible offline demo

Run a deterministic smoke demo over bundled source-code and StackOverflow fixtures:

python3 scripts/source_code_demo.py

The script prints token/line summaries and writes outputs/source_code_demo_summary.md. It intentionally avoids network scraping, BabelFish, and notebook execution.