Skip to main content
A model is a trained representation of your catalog and user behaviour. NeuronSearchLab trains models on the events you send and the catalog items you have ingested. Once a model is trained and approved, you can deploy it to the live endpoint and it will immediately begin serving recommendations.

​
Model families

Models are grouped into families. Each family represents an ongoing sequence of training runs on your data. Within a family, each new training run produces a new version. The console shows all families available to your account, including shared base models and models trained specifically on your tenant’s data. For each family the console shows:
  • The current version in production (if any)
  • The approval status of each version (Pending, Approved)
  • Creation date and training metadata

​
Approve a model

After a training run completes successfully, the new model package has a status of Pending. You must approve it before it can be deployed.
  1. Open Console > Training Jobs or Console > Models.
  2. Find the completed training run and click Approve.
  3. The model status updates to Approved.
Approving a model does not automatically change what is served. It makes the model available for promotion.

​
Promote a model to production

Once approved, you can promote a model to replace the current live endpoint.
  1. Find the approved model version.
  2. Click Promote to production.
  3. The console will show the endpoint status updating. Depending on infrastructure state this may take a moment.
  4. Once the endpoint status shows as InService, the promoted model is live and all subsequent recommendation requests will use it.
You can promote a previous version at any time if you need to roll back.

​
The production banner

The Models page shows a banner indicating which model version is currently serving recommendations. The banner includes:
  • The model name and version number
  • The endpoint status (InService, Updating, etc.)
If no model is currently deployed the banner is not shown and recommendation requests will return empty results.

​
Training run metrics

Each training run records a set of metrics at completion. Open a run from the Training Jobs page to view:
  • Training job status and duration
  • Final metrics (loss, accuracy, or custom metrics your training job reports)
  • The run manifest, which records the configuration used
Use these metrics to compare runs and decide whether a new version is ready to promote.

​
Trigger a new training run

See the Events and Signal Templates guide for instructions on configuring event types, setting thresholds, and starting a training run.