Approving deployed models
Before the model within an ML deployment can generate predictions, it needs to be activated by someone with sufficient permissions. When a model is activated, all ML deployments that use this model are activated for predictions.
Models can be activated and deactivated as needed to help your organization optimize the usage of the Qlik Cloud subscription. The process of activating and deactivating models is known as model approval.
Guide for administrators
This help topic describes the process of model approval for users who do not have any administrator privileges. For information about model approval for administrators, including assignment of model approver permissions, see Working with model approval as an administrator.
About model approval
Model approval allows both users and administrators to control the number of deployed models within the Qlik Cloud subscription that are activated for making predictions. The Qlik Cloud subscription limits the number of deployed models that can be used for predictions at a given time. With model approval, users and administrators can activate or deactivate models, and assign permissions for model approval accordingly. This allows a more efficient usage of the subscription.
Model approval status
The approval status of a deployed model indicates whether it can be used to make predictions. A deployed model can be in one of the following statuses at a given time:
-
Active: The model is activated and able to generate predictions.
-
Inactive: The model is deactivated and cannot generate predictions.
-
Requested: Approval for the model has been requested but not yet provided. When a model is in Requested status, it does not count towards the total number of deployed models allowed for the subscription.
Model approval status shown at the top of the interface when you open an ML deployment.
Model approval status shown at the top of the interface when you open an ML deployment. In this case, the model in the deployment is Active, meaning it can be used for predictions.
Activating and deactivating a model as a user
Do the following:
-
Open an ML deployment.
-
At the top of the interface, use the toggle switch to change the model to Active or Inactive.
For the equivalent procedure for administrators, see Activating and deactivating models as an administrator.
How to know if you can activate and deactivate models
Your ability to approve and reject models is dependent on permissions assigned to you by tenant administrators. For specifics, see Methods and requirements.
If you are able to activate and deactivate a model for predictions, you can see toggle switch at the top of the interface when you open the ML deployment. This toggle switch allows you to activate or deactivate predictions from the source model. You can also view the subscription's current capacity for active models.
Model approval status shown at the top of the ML deployment interface, with toggle switch available.
If you are not able to activate or deactivate a model for making predictions, you can still see the approval status of the source model at the top of the interface. However, you cannot change this status.
Model approval status shown at the top of the ML deployment interface, with toggle switch not available.
Use cases
The primary benefit of model approval is to allow more efficient usage of the Qlik Cloud subscription. If desired, your organization can additionally use model approval to ensure optimum quality in deployed models. For example, you can assign a select number of users with permissions for model approval and restrict model approval for others. These model approvers can enforce quality assurance to help your organization generate more accurate and reliable predictions.
Deactivating a deployed model is an alternative to deleting the ML deployments that use it. When you deactivate a model, it is not counted as a usable deployed model in your Qlik Cloud subscription. This allows you to more efficiently use your tenant's allotted capacity for deployed models. For example, you can deactivate outdated models until new training data becomes available, allowing other users to run predictions from other models while there is capacity to do so.
Methods and requirements
The model within an ML deployment needs to be approved before it can generate predictions. Model approval can be performed by users and administrators.
| Approval method | Where approval is performed | Required permissions |
|---|---|---|
| User | ML deployment |
All of the following:
|
| Administrator | Administration activity center |
One of the following:
|
User method for model approval
A user (non-administrator) opens the ML deployment, and then activates or deactivates the source model.
Administrator method for model approval
An administrator activates or deactivates models from the AutoML section in the Administration activity center.
Model approval permissions
This section outlines the permissions needed to activate and deactivate models. Model approval permissions are controlled through edits to the User Default role and assignment of custom roles.
For more information about how to assign permissions, see:
-
Roles and permissions for users and administrators (user-based subscriptions)
-
Roles and permissions for users and administrators (capacity-based subscriptions)
For more information about the available permissions in Qlik Cloud, see:
-
Permissions in User Default and custom roles (user-based subscriptions)
-
Permissions in User Default and custom roles (capacity-based subscriptions)
User model approver permissions
The Approve and reject your AutoML models permission controls a user's ability to activate and deactivate models from the ML deployment interface. The Approve and reject AutoML models administrator permission also controls this ability.
Space roles further control a user's ability to perform model approval actions. To activate and deactivate models from an ML deployment, the user needs to have:
-
Shared spaces: Owner, Can manage, or Can edit role in the space where the ML deployment is located.
-
Managed spaces: Owner or Can manage role in the space where the ML deployment is located.
Administrator model approvers can activate and deactivate models from any ML deployment to which they have the applicable space roles listed above. To activate and deactivate models used in an ML deployment within another user's personal space, an administrator model approver needs to use the Administration activity center.
Administrator model approver permissions
The Approve or reject AutoML models administrator permission, with a value of Allowed, gives the user the ability to activate and deactivate models. This permission allows you to perform these actions from the Administration activity center, and additionally from the ML deployment interface (if you have corresponding space roles).
All tenant administrators can also activate and deactivate any model from the Administration activity center.
Workflow
This section outlines a sample workflow that can be used to make the most out of model approval.
Step 1: A user deploys the model
When a model is first deployed into an ML deployment, it enters Requested approval status. This means that the model becomes visible in the Administration activity center and administrators can approve or reject it. Any user who opens the ML deployment will also receive a notice letting them know the model needs to be approved before it can generate predictions.
Step 2: The model is approved or rejected
A user or administrator needs to activate the model for making predictions. If a decision is made to not allow the model to generate predictions, an administrator approver can set it to Inactive, or it can be left in Requested status until approval is needed.
When the source model in an ML deployment is approved and able to generate predictions, it has the Active status. When the source model in an ML deployment has been deactivated from being able to make predictions, it has the Inactive status.
Step 3: The model status is changed over time
Over time, a model might be replaced by other models, or the tenant might need to deactivate some models to allow other models to make predictions. This can be a temporary solution to more efficiently use the deployed models included in the subscription, or it might be permanent if you want to keep the model for reference purposes only.