ASE2021 vulnerability fix dataset
Bibliographic record
Abstract
The dataset of "Finding A Needle in a Haystack: Automated Mining of Silent Vulnerability Fixes", which was accepted in the 36th IEEE/ACM Automated Software Engineering (ASE) Conference. Followings are the descriptions of columns: commit_id: The commit ID/hash. repo: The Github Author and repository (e.g., "apache/hive"). filename: The name of the file changed in the commit. partition: Which dataset the commit information belongs to (i.e., "train", "val", or "test"). PL: Programming Language (PL) (i.e., "java" or "py"). label: Label of the commit, 0 for non-vulnerability fixing commit and 1 for vulnerability fixing commit. diff: The entire code change information of the file in this commit. committer_date: The date of the commit (e.g., 2015-03-02 13:48:25+13:00) msg: The commit message (NA if empty). MOD_DIFF: The code change of the file in this commit after preprocessing: filtering out lines that are not added lines or removed lines, and removing refactoring information and comments. BPE_MOD_DIFF: BPE processing applied to MOD_DIFF information (using codeprep Python package). ADD_DIFF: The added lines from the MOD_DIFF information (indicated as a line starting with '+' character). REM_DIFF: The removed lines from the MOD_DIFF information (indicated as a line starting with '-' character). LOC_ADD: Total lines of code added in this file change. LOC_REM: Total lines of code removed in this file change. LOC_MOD: Total lines of code modified in this file change (LOC_ADD + LOC_REM). commit_repo: The commit ID and repository concatenated. cve_list: A list of CVEs which the commit fixes (e.g., CVE-2015-5348, CVE-2016-8902). Following is the code snippet to reproduce Table 1. import pandas as pd all_commits = pd.read_csv('./ase_dataset_sept_19_2021.csv') #Separate by language, since the Java commits are missing some info which we will add later on. py = all_commits[all_commits.PL == 'python'] java = all_commits[all_commits.PL == 'java'] #Java first: partition into train/val/test and check # of commits print("Java VF vs NVF for train/val/test") java_train = java[java.partition =="train"] java_val = java[java.partition == "val"] java_test = java[java.partition == "test"] print(java_train.drop_duplicates(subset='commit_id').label.value_counts()) print(java_val.drop_duplicates(subset='commit_id').label.value_counts()) print(java_test.drop_duplicates(subset='commit_id').label.value_counts()) #Python: partition into train/val/test and check # of commits print("Py VF vs NVF for train/val/test") py_train = py[py.partition =="train"] py_val = py[py.partition == "val"] py_test = py[py.partition == "test"] print(py_train.drop_duplicates(subset='commit_id').label.value_counts()) print(py_val.drop_duplicates(subset='commit_id').label.value_counts()) print(py_test.drop_duplicates(subset='commit_id').label.value_counts())
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.010 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.007 | 0.005 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.003 | 0.002 |
| Insufficient payload (model declined to judge) | 0.014 | 0.022 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".