0 / 0
Registering models for tracking
Registering models for tracking

Registering models for tracking

You can register models from a project, space, or catalog so that the details for the model lifecycle display in the model inventory for a model entry. After you register the model, details are recorded in the associated factsheet.

Registering a model

When a model is registered to a model entry, the factsheet service creates a relationship between the model and the model entry. You can view the connection in the Related assets section of model entry. Note the following relationships:

  • a model entry tracks a machine learning model
  • the machine learning model is registered to the associated model entry

Note: When a model is removed from a model entry, the relationship is removed from the Related assets section. You can manually delete the relationship from the Related assets section, but it is restored if the model is still registered to the model entry.

Registering a model from the model details page of a project, space, or catalog:

  1. Open the Models page for in a project, space, or catalog.
  2. Select a model entry and click Track this model.
  3. Choose to associate the model with an existing model entry in the inventory or create a new entry.
  4. Specify the details for the new entry, including specifying a catalog if you have access to more than one, and save to register the model.
  5. A link to the model inventory is added to the model details page. Click the link to open the model entry in the inventory.
  6. Optionally update the model entry. For example, add tags, supporting documentation, or other details.

Registering a model in a notebook

If your model is in a notebook, you can insert code similar to this sample code at the end of the notebook to register the model with a model entry:

props = ModelEntryProps(
           model_entry_catalog_id='ENTER CATALOG ID TO CREATE MODEL ENTRY',
           model_entry_name="notebook-entry",
           model_entry_desc="Created from notebook",
           asset_id='Enter your Model asset id here',
           project_id='Enter your project id of your model here'
           )

registration_details = client1.register_model_entry(model_entry_props=props)

For details on using Python commands to interact with the model inventory, see Python client documentation and the IBM_AIGOV_FACTS_CLIENT documentation.

Removing a model from an entry (unregistering)

Click Deactivate to disassociate the model from the model entry.

Viewing all registered models for a model entry

A model entry can include multiple registered models. For example, you can have a deployed model and a challenger model. If the challenger model performs better, you might want to patch the deployment and replace the deployed model with the better-performing one.

To view all registered models for a model entry

  1. Open the model entry and click the Asset tab.
  2. In the Asset type column, select Model.

You can view the model and its facts for various containers, such as a project, pre-production deployment space, and production deployment space. Challenger models display on their own row. From this view you can see which model version was used to create a deployment, the status of a deployment (for example: Pending evaluation, Evaluated, Approved), and whether any alerts signal the need for attention.

Learn more

Learn about viewing model entries and factsheets.

Parent topic: Managing an inventory of model assets.