DRV Project (Defect-Risk Visualization)

OVERVIEW

DRV (Defect-Risk Visualization) is a framework that helps developers see and prioritize defect-prone code modules. It combines defect prediction models with interactive visualizations so teams can quickly identify which parts of the codebase are most risky.

Objective

-Integrate software defect prediction (metrics + semantic models) into a unified workflow.

-Provide visual analytics (heatmaps, rankings, dashboards) to make predictions transparent and interpretable.

-Support developers in prioritizing high-risk modules, rather than treating all code equally.

-Deliver a practical tool that complements existing defect prediction research with developer-friendly insights.

This is a self-contained Python scaffold for DRV thesis project. It includes:

• ApacheJIT_Total Dataset
• Baseline models (metrics-only and fused semantic+metrics)
• Ranking metrics (FPA, Recall@K, NDCG@K)
• Visualization (risk heatmap, Top-K bar chart)
• CLI entrypoint

Note: Semantic representation uses TF‑IDF of commit messages as a stand-in for BiCC-BERT embeddings (offline-friendly). Replace with your BiCC module if available.

Quickstart

# From this folder:
python -m drv.cli --commits data/commits.csv --mode train --output outputs

Artifacts:

• outputs/train_metrics.json — F1/AUC for metrics-only and fused models
• outputs/ranking_eval.json — FPA/Recall@K/NDCG@K
• outputs/scored_commits.csv — risk scores for each commit
• outputs/heatmap.png, outputs/topk.png — example visualizations

Structure

• drv/config.py — config dataclass
• drv/data.py — CSV loaders/savers
• drv/features.py — CK feature handling + fusion
• drv/models.py — training and scoring (GradientBoosting as a local stand-in)
• drv/ranking.py — FPA, Recall@K, NDCG@K
• drv/eval.py — F1, AUC helpers
• drv/visualize.py — heatmap & top-k plot
• drv/cli.py — command-line workflow