VFDelta Vulnerability Fix Dataset

The vulnerability fix dataset of VFDelta contains two datasets used in our study: VFM_2021 and VFM_2023. Refer to our paper for the detail of dataset. We are working on model publishing; please always check for the newer version.

vfm_2021_train_val.parquet

vfm_2021_test.parquet

vfm_2023_train_val.parquet

vfm_2023_test.parquet

eval_vfdelta.ipynb

The commit-level dataset provides meta information of "repo_name", "commit_id", "label". train_val file contains "partition" for the split of training and validation set, and test file contains "vfdelta_predict_proba" which is the predict probability of each commit by VFDelta. 

To evaluate VFDelta's performance on vfm_2021 and vfm_2023, run eval_vfdelta.ipynb, use the column vfdelta_predict_proba as our prediction result. You can also use them to compare the performance of your prediction results with VFDelta's.