Retraining the model of an object with SQL DI web UI

Enabling an object for AI query trains a neural network model based on the data you select, and SQL Data Insights (SQL DI) uses the model to discover hidden relationships in the data and generate meaningful insights. As the data change, the accuracy of the model may degrade over time. When needed, you can retrain the model if it's eligible. You can retrain the model on the AI objects page of the SQL DI user interface.

Before you begin

Not all models are retraining eligible by default. A model is eligible for retraining only after your SQL DI environment is updated with required maintenances and the corresponding object is created and enabled for AI query in the updated environment. See Preparing for SQL DI installation and Installing, configuring, and starting SQL DI for details on required maintenances. See Step 4 below for more information about model retraining eligibility.

Model retraining uses the entire data set in the Db2 source table or view that you selected for the AI object. The data set includes both the original data and the delta. Model retraining also uses the existing column configuration that you created for enabling the object for AI query.

Retraining a model does not impact the model that is currently in use or disrupting the normal operation of your SQL DI. The current model is in use until you deploy the retrained model. See Deploying the retrained model of an object with SQL DI web UI for details.

Make sure that you have the authorization that is required for a specific task in SQL DI. See Managing SQL DI user permissions and Configuring setup user ID for SQL DI for details.

Procedure

Sign in your SQL DI web UI with a valid RACF® user ID at the following address:

https://<SQLDI-IPAddress>:<SQLDI-PortNumber>

SQL DI uses a login group to identify and authorize users. The default group name is SQLDIGRP. Make sure that the user ID that you specify is defined in your SQL DI login group.
On the Connections page, select a connection and click the action menu to display the task options.
Select List AI objects to open the AI objects page for the connection.
Select an object and click the arrow to expand the row and displays the Model retraining tile.

The Model retraining tile shows whether the object's model is retraining eligible. If the model is eligible, the tile also shows the status and timestamp of the last retraining job.

For a model to be eligible for retraining, you must first apply the required APARs/PTFs. When you create a new object in the updated SQL DI environment and enable it for AI query, the resulting model is by default eligible for retraining.

For an existing model, you must disable the corresponding object for AI query and then enable it again in the updated SQL DI environment. The re-enablement process trains a new model for the object. The new model is automatically eligible for retraining. See Enabling or disabling AI query with SQL DI web UI for instructions.
From the action menu, select Start retraining and then click Retrain to confirm.

SQL DI starts the retraining process in the background with the Retraining in progress status on both the AI objects page and the Model retraining tile. The entire process may take some time to complete depending on the size of the data set and the column configuration. You can monitor the progress by refreshing the page. If needed, you can terminate the retraining by clicking Stop retraining from the action menu.

Successful completion of the retraining process changes the status to Retrained with a green check mark and an updated timestamp. If the status is Failed, review the error message, fix any error, and restart the retraining process.

The retrained model is not automatically put into use. You must deploy the retrained model to use it. See Deploying the retrained model of an object with SQL DI web UI for instructions. If needed, you can delete the retrained model. See Removing a retrained model for instructions.