Hi PatchFinder authors,
Thank you for releasing the PatchFinder code and sample data. I am working on a CVE-to-fixing-commit retrieval/reranking study and would like to reproduce the full PatchFinder experimental setting.
From the repository documentation, I understand that the full dataset contains approximately:
4,789 CVEs
4,870 security patch commits
532 OSS projects
21,781,044 total candidate commits after adding same-repository non-patch commits
train/validation/test splits by CVE in an 80/10/10 ratio
The repository currently includes data/patch_data_4870.csv and sample split files, but the full train_data.csv, validate_data.csv, and test_data.csv are not included because of their size.
Would it be possible to share one of the following?
A download link for the full prepared dataset or compressed split files.
The exact CVE split lists used in the paper.
The scripts/configuration used to sample the same-repository negative commits.
Any cached commit metadata or NVD CVE description files used in the experiments.
If the full 110GB dataset is difficult to host publicly, even the CVE split files plus reconstruction scripts would be very helpful for reproducibility.
Thanks again for making PatchFinder available.
Hi PatchFinder authors,
Thank you for releasing the PatchFinder code and sample data. I am working on a CVE-to-fixing-commit retrieval/reranking study and would like to reproduce the full PatchFinder experimental setting.
From the repository documentation, I understand that the full dataset contains approximately:
4,789 CVEs
4,870 security patch commits
532 OSS projects
21,781,044 total candidate commits after adding same-repository non-patch commits
train/validation/test splits by CVE in an 80/10/10 ratio
The repository currently includes data/patch_data_4870.csv and sample split files, but the full train_data.csv, validate_data.csv, and test_data.csv are not included because of their size.
Would it be possible to share one of the following?
A download link for the full prepared dataset or compressed split files.
The exact CVE split lists used in the paper.
The scripts/configuration used to sample the same-repository negative commits.
Any cached commit metadata or NVD CVE description files used in the experiments.
If the full 110GB dataset is difficult to host publicly, even the CVE split files plus reconstruction scripts would be very helpful for reproducibility.
Thanks again for making PatchFinder available.