How hpci-lab/yprov4ml_mlflow_plugin works
Project type and purpose
• This is an open-source MLflow plugin (“yProv4ML_MLflow_Plugin”) that provides a custom storage backend (in store.py) to capture and persist ML provenance metadata alongside your regular MLflow artifacts.
• Key indicators: a single Python package directory (mlflow-yprov4ml-plugin), a store.py implementing a store interface, and a setup.py for Python packaging.
File structure
• Top level:
• .gitignore, LICENSE, README.md – standard project metadata.
• setup.py – defines entry points to register the plugin with MLflow.
• mlflow-yprov4ml-plugin/ – contains store.py, the heart of the plugin.
• Patterns: This is a plugin targeting MLflow’s plugin architecture (not a full‐stack app).
• No “frontend” or “backend” folders, but an extension point: MLflow’s store interface.
README for additional insights
• The README only gives the plugin name, implying that the detailed documentation is either within code comments or external docs.
• We infer from the package name and structure that the plugin registers a new store for MLflow Tracking (replacing or augmenting MLflow’s default artifact/metadata store) to capture provenance.
Elements
• Clear labels: “MLflow Client”, “MLflow Tracking Server”, “yProv4ML Store Plugin”, “Underlying MLflow Storage”, “Provenance Database”.
• Client → Tracking Server API
• Tracking Server → Plugin
• Plugin → Provenance DB (for metadata)
• Plugin → Underlying Storage (for artifacts/metrics)
• Blue rectangles for MLflow core components.
• Green for the yProv4ML plugin.
• Grey rectangle for the underlying storage (file system or SQL).
Final checklist for the engineer
• Plugin’s dual‐write behavior (artifact vs. provenance) is visually distinguished.
• Annotations note the MLflow entry point in setup.py and the main logic location (store.py).