driftwatch 0.2.0

Lightweight ML drift monitoring, built for real-world pipelines

DriftWatch

Lightweight ML drift monitoring, built for real-world pipelines

---

📖 Documentation

Read the full documentation here: vincentcotella.github.io/DriftWatch

🚀 Features

• Simple API: Detect drift in 3 lines of code.
• Multiple Detectors: PSI, KS-Test, Wasserstein Distance, Chi-Squared.
• Production-Ready:
  • ⚡ FastAPI Integration (Middleware included).
  • 🔔 Slack Alerts built-in.
  • 🛠️ CLI for batch processing.
• Lightweight: Minimal dependencies (numpy, pandas, scipy).
• Type-Safe: 100% typed code with mypy support.

📦 Installation

pip install driftwatch

For extras (CLI, FastAPI, Alerting):

pip install driftwatch[all]

⚡ Quick Start

from driftwatch import Monitor
import pandas as pd

# 1. Initialize monitor with reference data (e.g., training set)
monitor = Monitor(reference_data=pd.read_parquet("train.parquet"))

# 2. Check production data for drift
report = monitor.check(pd.read_parquet("production.parquet"))

# 3. Act on results
if report.has_drift():
    print(f"⚠️ Drift detected! Ratio: {report.drift_ratio():.1%}")
    print(f"Drifted features: {report.drifted_features()}")
else:
    print("✅ All systems normal.")

🛠️ Usage Scenarios

Scenario	Solution
Real-time API	Use DriftMiddleware in FastAPI to monitor every request.
Batch Job	Use driftwatch check CLI in your Airflow/Cron jobs.
CI/CD	Block deployments if validation data drifts from training data.
Alerting	Send Slack notifications automatically when drift is critical.

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

1. Fork the repo.
2. Install dev dependencies: pip install -e ".[dev,all]"
3. Run tests: pytest
4. Submit a PR!

📄 License

MIT © Vincent Cotella