Migrating assets

Starting on September 1, 2020, new features are available on IBM Watson Machine Learning, including:

  • Deployment spaces for creating, monitoring, and running deployments
  • Machine learning assets are stored in Cloud Object Storage rather than in the Watson Machine Learning repository
  • Support for the v4 version of the Watson Machine Learning Python client library
  • Import or export of spaces and assets
  • Support for a broader set of input data sources for batch deployments
  • Updated service plans for building and deploying machine learning models

To take advantage of new features, you must migrate assets you want to retain to a project before you upgrade your Watson Machine Learning service instance. A migration tool makes it easy for you to move your assets.

You can migrate:

  • Experiments
  • Functions
  • Models
  • Pipelines

You must create new deployments after you migrate assets and upgrade your service instance.

Attention: At the end of the designated migration period, assets that were not moved will be deleted. For details on the dates of the migration period, see What’s New.

Launching the migration tool

You will be prompted to migrate assets when you access Watson Machine Learning using a deprecated service instance. Follow these steps to migrate assets and then upgrade your Watson Machine Learning service instance.

  1. In the Watson Machine Learning section of your project assets, a banner displays with information on the migration. Click Migrate now to start the migration process.
  2. Select the assets to migrate. You do not have to migrate all assets. You can continue to migrate assets until the end of the migration period. Note: when you choose the assets to migrate, some, like pipelines, will not display as project assets but will be available by API. Some dependent assets, such as software specs, are required and will display in the project after migration even though you can’t select them here.
  3. (Optional) You can filter by asset type if you want to migrate all models, for example. You can also search for assets by name.
  4. Click Migrate to start the migration. You do not have to remain on the page while the migration process runs. A confirmation message informs you when the assets are successfully migrated to the project. A migration summary shows what you migrated, whether there were any errors, and what assets remain. A migration log is available to help you understand any errors.
  5. When you are finished migrating assets, open the Settings tab for your project to upgrade your Watson Machine Learning service instance.

Watch this short video to see how to migrate assets to use the upgraded Watson Machine Learning service.

Figure 1. Video icon Migrate assets to use the upgraded service.

SPSS Modeler flows might require retraining

After migrating your assets and associating a v2 machine learning service instance, you might have to retrain your SPSS Modeler flows and save them to Watson Machine Learning. Retrain if a model uses any of the folowing 6 nodes before deploying them or new deployments will fail.

  • XGBoost Tree
  • XGboost Linear
  • One-Class SVM
  • HDBScan
  • KDE Modeling
  • Gaussian Mixture

For existing deployments created using the v3 or v4-beta deployment service APIs and containing any of the 6 nodes, you can continue to score the deployments for one month, Until October 1, 2020.

Migrating assets using the Watson Machine Learning API

If you prefer to migrate assets programmatically, use a dedicated set of migration APIs to move your machine learning assets to a space or a project before you associate a new Watson Machine Learning service instance. Refer to the API documentation for migration instructions and examples.

Migrating assets using the Watson Machine Learning Python client

For examples of using the Python client to migrate assets from the v3 Python client library or from the v4 beta client, see these sample notebooks:

Next steps