By Jeren Mckey, Db2 for z/OS Development, and Guanjun Cai, Db2 for z/OS Content Design
Db2 13 for z/OS (Db2) introduced the SQL Data Insights (SQL DI) feature at its GA. The feature infuses artificial intelligence (AI) into the Db2 engine and enables AI queries for discovering valuable insights in your data. Based on your feedback, Db2 has continuously delivered new and enhanced functions for the feature as they become available.
The latest update to the SQL DI feature is the model retraining capability.
When you enable an object for AI queries, SQL DI takes a snapshot of the data that you select from a source Db2 table or view and trains a neural network model based on the static selection. But your data are dynamic and changing. As your data change, the accuracy of the trained model may degrade over time. To ensure its accuracy, you need to retrain the model.
Currently, to retrain the model of an AI object, you must disable the object for queries, which deletes the model, and enable the object for queries again. This process can be disruptive depending on the data size and the retraining frequency. Once the model is deleted, you cannot run AI queries or execute AI functions until the re-enablement is complete. The larger the data set gets, the longer it takes to train a new model. The more often you need to retrain the model, the more retraining time you will need to factor into your operations.
The new model retraining capability will change that. With this update, you can retrain the model while it is still in use. In other words, retraining the model does not interrupt your normal execution of AI queries and AI functions. When you initiate the retraining process, SQL DI trains a new model against the entire data set, including the new and changed data, and leave the existing model untouched. After retraining completes successfully, you can choose to deploy the new model at any time. The deployment will remove the existing model and put the new model into active use. Optionally, you can choose to delete the new model if you don’t plan to use it.
See “Running AI queries with SQL Data Insights” to learn more about Db2 SQL DI and its new model retraining capability.
A special thank-you to Akiko Hoshikawa (Distinguished Engineer, Db2 for z/OS and AI Product Development), Jae Lee (Db2 for z/OS Development), and Jason Cu (Db2 for z/OS Development) for their contributions.