Creating and working with project groups

Project groups allow you to group trained models with the data sets that were used for training.

This grouping is optional but is a useful way to organize related data sets. For example, project groups can be useful with a workflow that clones data sets as you refine labels and work toward a more accurate model. Project groups can be used with a production work flow strategy and automatic model deployment for even more functionality.

Project groups provide API shortcuts for certain trained model actions. That is, you can deploy or perform inferences on the most recently trained or deployed model without knowing the model ID. Instead, the APIs use project group IDs, which never change. As better performing models are generated, your scripts can act on the latest model without needing to be updated.

Project groups track the latest trained model and the latest deployed model separately. If production work flow is enabled, you can additionally add tags to denote a trained model and its deployed instance as production-ready or as untested. See Production work flow for more information about these tags. When using project groups with the production work flow, you can use project group APIs to work with the following models:

  • Latest trained model in a project group.
  • Latest deployed model in a project group.
  • Latest trained model that is production-ready in a project group. Production work flow must be enabled for the project group.
  • Latest deployed model that is production-ready in a project group. Production work flow must be enabled for the project group.
  • Latest trained model that is untested in a project group. Production work flow must be enabled for the project group.
  • Latest deployed model that is untested in a project group. Production work flow must be enabled for the project group.
Note: All models trained from any data set in a project group are automatically associated with that project group.

Working with project groups and project group assets

Project groups can be created at any point in your work flow. To create a project group, click Projects in the side bar, then click +. After the project group is created, you can add resources, such as data sets and trained models, to it.

To delete a project group, from the Projects page, select the project name and click the trash can icon. None of the assets in the project are deleted, but they are no longer associated with any project group.

Working with project group assets

To work with project group assets, navigate to the Projects page and click the name of the project group.

Adding an asset

To add a data set or model, click +, specify the asset type, and select the asset to add. You can start typing the asset name to filter the available assets.

An administrator can add assets to a project group that was created by a different user. However, the project group owner will not be able to see the added assets because only administrators can see resources that were created by other users. The value for "Total items" on the Projects page might be larger than the number of items shown on a project's details page.

Removing an asset

To remove an asset, navigate to the project group, select all assets that you want to remove, and click the remove icon (-). The assets are not deleted from IBM® Maximo® Visual Inspection. Additionally, each asset in a project group is independent. For example, if you remove a data set, none of the models derived from that data set are removed.

Notes

  • Any model trained from a data set in a project group is automatically added to that project group. However, any models that were trained from a data set before it was added to the project group must be added manually.
  • Each data set or trained model can be a member of only one project group.

Using the production work flow with project groups

If the production work flow is enabled, project groups keep track of the most recently trained model that is marked Production and the most recently trained model that is unmarked. You can use an API to work with the latest deployed model that is marked Production or is Unmarked (untested). This feature simplifies your workflow because you never have to update the script to point to a different deployed model, and you do not have to manually track model names.

Note: Because the latest trained model is tracked separately from the latest deployed model, it is possible to train a new model and still be using an older model for inferences. This delineation can be reduced if you enable the production work flow and auto deploy. With both of these flags set, the project group tries to keep the deployed models in sync with the trained model's latest trackers.

For details, see Production work flow.

Automatically deploying models in project groups

If you are using the production work flow within a project group, you can also turn on auto deploy. When auto deploy is turned on, IBM Maximo Visual Inspection automatically deploys a model when it is successfully trained and when it is marked as Production. IBM Maximo Visual Inspection automatically undeploys any deployed models when the associated trained model is marked as Rejected. Additionally, it tracks the latest model marked as Production or Untested and ensures that the latest production-ready model is deployed for a project group. For details, see Automatically deploying the newest model.