Vertex AI automatic retraining

Question

I'm trying to create a Vertex AI endpoint with Monitoring enabled that can trigger a Vertex AI pipeline execution when one of the deployed models drops its performance. However, Vertex AI does not provide any built-in feature to do it. Is there a method to capture the alert thrown by Vertex AI Monitoring and trigger the Pipeline?

Answer 1

1. The Vertex AI Model Monitoring jobs are logged as part of Cloud Logging 1 .
2. You can react to those loggings using log-based alerts2 . For that, you need to configure a notification channel to PubSub 3
3. Based on those PubSub message you can trigger a Cloud Function 4
4. The Cloud Function can initiate the Vertex AI Pipeline run to re-train the model 5

https://cloud.google.com/vertex-ai/docs/model-monitoring/using-model-monitoring#cloud-logging-info

https://cloud.google.com/logging/docs/alerting/log-based-alerts

https://cloud.google.com/monitoring/api/ref_v3/rest/v3/projects.notificationChannels#NotificationChannel

https://cloud.google.com/functions/docs/calling/pubsub

https://cloud.google.com/vertex-ai/docs/reference/rest/v1/projects.locations.pipelineJobs/create