In src you can find the code for the RepositoryDataScraper. In main.py you can see how to use the class.
The repositories that we consider are sourced from SEART with the following filters:
- Last commit between: 01/01/2020 - 31/05/2024
- Branches ≥ 5
- Contributors ≥ 10
- Stars ≥ 100
- Exclude forks TRUE
- Has License TRUE
- Specifically one of: MIT License, Apache License 2.0, BSD 3-Clause "New" or "Revised" License, BSD 2-Clause "Simplified" License
We manually download the metadata from SEART and then programmatically clone
the repositories, before using RepositoryDataScraper to scrape them.
In doing so, we mine samples for the following agent scenarios:
- Subsequent commits modifying the same file (file-commit grams)
- Merging
- Cherry-picking
You can use the RepositoryDataScraper to scrape your own repository or you perhaps would like to extend it
to mine additional scenarios for AI Agents. In either case, follow the setup instructions in this section to
setup and run the system locally.
To simply run the tests, please first unzip the repositories for testing that we provide in repos/testing-repositories.zip
into the repos folder.
Some files are just included for documentation purposes, such as src/notebooks/analyze_dataset.ipynb for which
the raw dataset .csv is not included. We will probably release the dataset on HuggingFace at a later point.
This issue can occur on Windows, since the new maximum length for file names is opt-in. Please refer to this documentation on how to fix the issue. We recommend setting the git workaround, as we still had issue after updating the registry key and restarting.