Monitoring

Verta Model Monitoring enables Operations and Data Science teams to:

• Identify failing models before customers do

• Detect unexpected changes in model I/O including drift and outliers

• Detect model quality issues

• Resolve production issues with alerting & root cause analysis

What are the benefits of a model monitoring system?

• Your deployed models are automatically monitored with the system auto-configuring model predictions, features, alerts and thresholds with templated dashboards.

• Detect data drift and outliers in input features and predictions

• Monitor custom metrics and build your own charts and visualizations (e.g. confusion matrix, PR curve, ROC curve and more)

• Fully customizable dashboards and interactive charts

• Ingest ground truth (realtime and delayed) and monitor quality metrics like accuracy, precision, recall, F1 score etc

• Connect pre-production (model catalog and experiment tracking) and production systems for end to end visibility and root cause analysis

• Receive actionable alerts for performance degradation, or drift and automate remedial actions

• Track and compare model releases and compare performance between versions