GHPR Dataset

GHPR Dataset Overview

Dataset Description

• Name: GHPR Dataset
• Purpose: Empirical research and evaluation of software defect prediction
• Record Count: 3,026 defect‑fix records based on GitHub Pull Requests (PRs)
• Learning Instances: 6,052 total (3,026 defective, 3,026 non‑defective)

Data Format

• File Formats: Two formats are provided
  • ghprdata.csv: Compatible with Python's NumPy or pandas
  • ghprdata.sql: UTF‑8 encoded, suitable for large‑scale databases

Data Features

• Record Features: Each record contains 16 features, including project name, project owner, project description, project tags, programming language, pre‑ and post‑fix version IDs, defective code, commit description, commit time, pre‑ and post‑fix file contents, file‑path changes, PR title and description, etc.

Dataset Metrics

• Static Metrics: 21 static metrics are calculated for the 6,052 instances, such as coupling, method complexity, inheritance depth, response classes, method cohesion, etc., computed using the open‑source tool mauricioaniche/ck.

Citation Information

• Citation Requirement: When using this dataset in publications, please cite the following paper:
  • Authors: Jiaxi Xu, Fei Wang, Jun Ai
  • Journal: IEEE Transactions on Reliability
  • Title: Defect Prediction With Semantics and Context Features of Codes Based on Graph Representation Learning
  • Year: 2021