We're thrilled to announce new, powerful, fully customizable alerts!🚨🎉
Alerts help ML teams stay ahead of issues in business-critical models. However, alerts lose their value if they are inflexible and created without the user’s interests in mind; when not setup properly, alerts can become a hindrance, resulting in user fatigue from irrelevant or constant alert notifications.
At Fiddler, we empower data scientists and ML practitioners to resolve model issues quickly by alerting them to the most essential and highly-critical issues. Our new, powerful yet flexible alerts are fully customizable, enabling ML teams to have control over what and when they want to be alerted, so they can focus on the most important issues across models in both testing and production environments.
Alerts are critical to model monitoring for tracking the health of models throughout their lifecycle. They can be set to signal early warning on incidents like underperformance or model drift. In some scenarios even a small shift in data distribution can drastically impact model performance, demonstrating the value of time-critical alerts.
By tracking key metrics around model performance, model drift, data integrity, and traffic, users can quickly resolve issues and avoid damage to the business.
Users can specify how often they receive email or platform-native alerts using absolute or relative values. Additionally, they can specify the level of severity by assigning threshold values for metrics, so they can quickly identify and resolve high-priority issues first.
Now, through a single alerts dashboard, data scientists and ML practitioners can easily visualize all the alerts across any model or project, modify alert rules, and integrate with other alert services, such as PagerDuty! Users can customize the dashboard by applying additional filters or segmenting timeframes to identify critical alerts and quickly troubleshoot high-priority models that need immediate attention.
We are excited for customers to take advantage of our powerful and customizable alerts, enabling them to focus on the most business and model critical issues. Read our documentation on alerts to learn more.