Refine and test your ML solutions iteratively on a non-production instance, and then use update sets to export the changes to your production instance. This practice mitigates the risk of retraining solutions on your live production instance.

Before you begin

Prior to testing on a test instance, ensure that the instance hosts recently-cloned data from your production instance so that the solutions you train on the test instance remain valid when you export them to production.

Role required: admin or ml_admin

About this task

Plan your changes carefully, and remember that update sets match records based on the system ID [sys-id] and not the version number. Note that moving solutions to instances can sometimes deliver unpredictable results. If you run into such an issue, retrain the solution again, which takes 5 minutes per solution. For more information on update sets, see System update sets .

Procedure

  1. Navigate to All > Predictive Intelligence > Classification > Solution Definitions or Predictive Intelligence > Similarity > Solution Definitions.
  2. Click the name of your trained ML Solution Definition record to open it.
  3. In the Related Links section, click Add solutions to the current update set.
  4. Click Update.

Result

Your trained ML solution artifacts, such as solution definitions, template records, and predictive model statistics, are added to the current update set.

What to do next

Depending on where you are in your solution testing, schedule your update set for export to another non-production instance for further testing, or on to production.
Note: After you export a similarity solution, click Refresh similarity window (Required after Solution Import) in the Related Links section of the corresponding ML Solution Definition form.