Changing the deployed version of a model
Only one version of a model can be deployed at any one time. Where you have multiple versions of a model, you can change the deployed version.
Before you begin
You must have generated at least two versions of an AI model, and deployed one of them as described in Deploying models.
Procedure
-
In the AI Model Management, under Trainable AI algorithms, click the tile of the configuration you want to change.
-
When the overview page opens, click the Model versions tab. A list of model versions is displayed. For more information, see the List of model versions.
-
To change the deployed version, click the options menu at the far end of the table in the row corresponding to the version that you want to deploy. Then, click Deploy. The table updates to show that the version that you selected is now the deployed version.
Reference information
The following information is referenced in this task.
List of algorithms configurations
| Column | Description |
|---|---|
| Name | Name of the algorithm configuration. |
| Latest model version | Latest version of the model. A new version of the model is generated every that time the algorithm configuration is run. |
| Version deployed | This is the currently deployed version of the model. If training is scheduled and deployment is automatic, then the value in this field always matches the value in the Version field. |
| AI algorithm | The AI algorithm that is related to this record. Possible values are: Change risk, Log anomaly detection, Similar tickets, and Temporal grouping. |
| Data prechecked | Date and time this configuration was last prechecked. |
| Date trained | Date and time this configuration was last run. If it has never been run, then this field is blank. |
List of model versions
The following columns are common to most algorithms and then they can vary in terms of additional columns. For example, Log anomaly also has 'Totals resources found', and 'Models created', whereas Change risk also includes 'Closed tickets' and 'Incidence avoidance'.
| Column | Description |
|---|---|
| Model | Version of the model. A new version of the model is generated every that time the algorithm configuration is run. |
| Status | The deployed version is marked as Deployed with a comforting green tick to the left of the word Deployed. The other versions have the status Complete. |
| Last trained | Date and time when this version of the model was trained. |
| Duration | Length of time taken to train this version of the model. |
| Data Quality | Possible values are Good and Needs improvement. Only deploy model versions that have the status Good. |