How to move your Predictive Intelligence Model to another instance with the Trained Solution - Support and Troubleshooting

How to move your Predictive Intelligence Model to another instance with the Trained Solution Summary This KB will help walk you through how to move/transfer a Predictive Intelligence Model and the Trained Solution.

Instructions We have a few different methods for moving Predictive Intelligence models from one instance to another (Source to Target).

Add the Solution to the current Update Set. Create a Scoped Application and use an App Repo.

1. Add the Solution to the current Update Set.

You can add a Solution Definition along with the Solution and Word Corpus to an Update Set by going to the Predictive Intelligence Solution Definition and clicking " Add solutions to current update set " under the Related Links.

This feature will work to bring your Solution Definition and related records over to another instance via an Update Set; however, this can also fail if there is already a trained solution.

When adding the Solutions to an Update Set, it will include the ml_model_artifact record along with the sys_attachment record for the Trained Solution. This Trained Solution on sys_attachment can be rather large for an Update Set which can cause the Update Set to fail when committing it on the Target Instance. Please note : 15MB is the limit size for a ml_model_artifact record in an Update Set. You can determine the size of the model by querying the sys_attachment table

If the attachment for the ML Model Store record > 15MB, then the recommended approach when using the " Add solutions to current update set " Related Link would be to add the solutions to an Update Set, then remove the ml_model_artifact Update records which also contain the sys_attachment Update records.

After removing those updates from the Update Set, you may now commit the Update Set on a Target instance. To bring over the Trained Solution, you can go to those same ml_model_artifact record you removed from the Update Set and Export it as XML. You can then import this XML on the Target instance to have it create the ml_model_artifact and sys_attachment records.

Please note that the Trained Word Corpus is not brought over either. This can and will cause the model to fail the training if there isn't a train Word Corpus on the instance. After committing the Update Set above, you can change a system property to train the Word Corpus which will allow the model to Train properly.

You can see how to handle the Word Corpus training and the property here: KB0864744

Steps to follow for this Update Set Method:

Create an Update Set with a meaningful name and set it as your Current/Active Update Set. Go to your Solution Definition for the Predictive Intelligence Model you would like to copy from your Source to Target instance. Click the " Add solutions to current update set " link under Related Links. Go to the Update Set that the you created. Remove the ml_model_artifact records from the Updates. Move the Update Set to the Target instance via Export/Import, or through the Remote Update Set Retrieval. Preview and Commit the Update Set. Train the Word Corpus if needed (If a trained Word Corpus doesn't already exist on the instance). KB0864744 You now have two options to have a working Trained Solution on the Target instance.

You can Train the Solution directly on the Target instance. You can go to the ml_model_artifact record and Export it as XML and Import it into the Target instance.

Solution Definition:

Update Set: (Note the checkboxes are the records we need to remove)

Example of an ml_model_artifact record that we need to remove:

2. Create a Scoped Application and use an App Repo.

This approach is normally the best Long Term approach as it's easy to train the model using your specific datasets, then you can push the new solution to your APP Repo where you can then have the app update on your target instances.

This method will bring over the trained solution and handle everything needed to make sure your Target instance matches your Source instance. This can be very helpful when you need to make sure your predictions need to closely match, as training a Model on a different instance can result in different prediction scores.

Here are a couple of Docs that can help in creating the application:

Creating an Application: https://docs.servicenow.com/bundle/rome-application-development/page/build/app-engine-studio/concept/aes-overview.html

Source Control Overview: https://docs.servicenow.com/bundle/rome-application-development/page/build/applications/concept/c_SourceControlIntegration.html

Link your App Repo to the application in Studio via Source Control: https://docs.servicenow.com/bundle/rome-application-development/page/build/applications/task/t_LinkAnApplicationToSourceControl.html