Administering Qlik AutoML

Tenant administrators can administer AutoML resources and user permissions in the Administration activity center.

Administering AutoML involves the following:

• Viewing and administering ML resources and jobs, including activating and deactivating models for making predictions. This is configured in the AutoML section of the Administration activity center.

• Controlling user access and permissions for working with AutoML resources. This is configured in the Users section of the Administration activity center.

• Monitoring the consumption of the allotted ML resources for the subscription. Monitor consumption metrics in the Home and AutoML sections of the Administration activity center.

For information on how to create experiments and deployments, see Machine learning with Qlik AutoML.

Types of administrators for AutoML

The following users have the ability to administer Qlik AutoML from the Administration activity center:

• Tenant administrators: User with the Tenant Admin role.

• Analytics administrators: Users with the Analytics Admin role.

• Model approver administrators: Users who do not have any administrator roles, but who have been assigned specific administrator permissions via the User Default and custom roles. Specifically, they need the Approve and reject AutoML models permission set to Allowed.

The following table outlines what each administrator can do.

Navigating the AutoML section in the Administration activity center

Administer AutoML in the AutoML section in the Administration activity center. All types of AutoML administrators can view the information in this section. Depending on which type of administrator you are, you might be restricted from performing certain actions.

Deployed models

The Deployed models tab shows all the models that have been deployed into ML deployments. Administrators can manage the following:

• Activate and deactivate models for making predictions from associated ML deployments.

  Working with model approval as an administrator

• View the source ML experiment where a model was trained.

• View the approval status and last approver of a model.

• Monitor all instances of where a model is deployed.

Click next to a model to access additional details, including model history, details about the source experiment, and the name of the training dataset.

ML deployments

The ML deployments tab shows all ML deployment in the tenant. Details available include:

• Date when the source model was deployed into the ML deployment.

• Name, status, and last approver of the source model.

Click next to a model to access details about the source model for an ML deployment, including model history and information about the source experiment.

Jobs

In the Jobs tab, manage AutoML jobs. For more information, see Stopping or canceling jobs.

Settings

The Settings tab allows you to configure additional options for model approval notifications across the tenant. For more information, see Configuring an additional approval notice.

Managing permissions for model approval

To use a deployed model for making predictions, a user or administrator with the right permissions needs to activate it. As an administrator, you can control which users in the tenant are able to activate and deactivate models for making predictions. The process is different for user and administrator approvers. For more information, see:

• Configuring model approval permissions for administrators

• Configuring model approval permissions for users

Managing permissions for working with AutoML resources

For users to view and work with AutoML resources, the users need a combination of user entitlement and specific security roles. In shared and managed spaces, access controls are further defined by space roles.

To allow users to work with experiments, deploy models, and create predictions, assign the Automl Experiment Contributor or Automl Deployment Contributor security role, or both, to these users.

For more information, see:

• Who can work with Qlik AutoML

• User-based subscriptions: Assigning security roles and custom roles

• Capacity-based subscriptions: Assigning security roles and custom roles

Model approval for administrators

Before generating predictions, a user or an administrator must approve the model within an ML deployment.

For more information about model approval for administrators, see Working with model approval as an administrator.

Configuring an additional approval notice

Whenever a user opens an ML deployment that uses a model that is pending approval, a message appears to notify them that model approval has been requested. This message is also shown when a user creates the first ML deployment from a given model.

As an administrator, you can add an additional notice to appear with this message. To modify the content of this notice, you need one of the following:

• Tenant Admin security role

• The Approve or reject AutoML models administrator permission set to Allowed

Do the following:

1. In the Administration activity center, go to AutoML.

2. Open the Settings tab.

3. In the Additional notice field, type the additional notice you want to show to users.

Stopping or canceling jobs

In the Administration activity center, tenant and analytics administrators can view all content about AutoML jobs. They can see currently running and queued jobs for model training, deployment, and prediction generation. Filter the list on the job type and on the username.

These administrators can stop or cancel jobs as needed.

Do the following:

1. In Administration, go to AutoML.

2. Open the Jobs tab.

3. Click next to a job.

   Information noteAlternatively, select multiple jobs by clicking the rows for each job.

4. Click Cancel job.

5. Confirm in the Job cancellation dialog.

The jobs are canceled.

Monitoring AutoML consumption for the subscription

You can monitor how many deployed models are currently activated for creating predictions. In the Administration activity center, open the Home or AutoML section. The following charts show how much of the deployed model capacity (counting only active models) is remaining for the subscription:

• AutoML deployed models

• AutoML deployed models with predictions active

This information can also be viewed in the ML deployment interface by any user who opens the resource. The information is shown in the model status section at the top of the page.

A Qlik Cloud Analytics subscription defines a maximum number of deployed models that can be active at the same time (across all tenants within the subscription, for multi-tenant subscriptions). This consumption limit is defined per model, meaning that multiple ML deployments created from a single model count as a single deployed model. If you have reached the maximum number of active deployed models, you can do one of the following:

• Deactivate one or more currently active models to make room for new ones.

• Delete one or more existing deployed models to make room for new ones.

• If you need to have all current and future models activated at the same time, upgrade the subscription to a higher tier. For information about upgrade options, see the Qlik Cloud® Subscriptions product description.