The dataset is gathered on Sep. 17th 2020 from GitHub. It has clean and complete versions (from v0.7): The clean version has 5.1K type-checked Python repositories and 1.2M type annotations. The complete version has 5.2K Python repositories and 3.3M type annotations. The dataset's source files are type-checked using mypy (clean version). The dataset is also de-duplicated using the CD4Py tool. Check out the README.MD file for the description of the dataset. Notable changes to each version of the dataset are documented in CHANGELOG.md. The dataset's scripts and utilities are available on its GitHub repository.

{"references": ["A. Mir, E. Latoskinas and G. Gousios, \"ManyTypes4Py: A Benchmark Python Dataset for Machine Learning-Based Type Inference,\" in 2021 2021 IEEE/ACM 18th International Conference on Mining Software Repositories (MSR), 2021 pp. 585-589. doi: 10.1109/MSR52588.2021.00079"]}