Skip to content

Home

DriftWatch
Lightweight ML drift monitoring, built for real-world pipelines
CI PyPI version Python 3.9+ License: MIT Coverage

What is DriftWatch?

DriftWatch is a production-ready drift detection library designed for ML engineers who need reliable, actionable drift monitoring without the overhead of complex platforms.

Key Features

  • Simple API

    Get started in 3 lines of code. No complex configuration required.

  • Production-Ready

    Includes FastAPI middleware, Slack alerts, and comprehensive CLI tools.

  • Multiple Detectors

    PSI, KS-Test, Wasserstein Distance, and Chi-Squared tests included.

  • Reliable

    Fully type-hinted, tested (96%+ coverage), and minimally dependent.

Quick Example

from driftwatch import Monitor
import pandas as pd

# Load your data
train_df = pd.read_parquet("train.parquet")
prod_df = pd.read_parquet("production.parquet")

# Create monitor
monitor = Monitor(reference_data=train_df)

# Check for drift
report = monitor.check(prod_df)

if report.has_drift():
    print(f"⚠️ Drift detected in {len(report.drifted_features())} features!")
    print(f"Drift ratio: {report.drift_ratio():.1%}")

Use Cases

1. API Monitoring

Monitor your FastAPI ML service automatically:

from fastapi import FastAPI
from driftwatch.integrations.fastapi import DriftMiddleware

app = FastAPI()
app.add_middleware(DriftMiddleware, monitor=monitor)

Read FastAPI Guide →

2. CI/CD Integration

Check for drift in your deployment pipeline to prevent bad rollouts.

driftwatch check --ref train.parquet --prod new_data.parquet

Read CLI Guide →

3. Slack Alerting

Get notified instantly when critical features drift.

if report.has_drift():
    alerter.send(report)

Read Slack Guide →

What's Next?

  • Get Started

    Run your first drift check in minutes

    Quickstart

  • Core Concepts

    Understand the math behind drift

    Concepts

  • API Reference

    Explore the full API documentation

    API Docs