Create a Criteria Set for a trained model, then define constraints for input features
and targets for output features. Send a completed Criteria Set and its model to the engine for
evaluation.
When you first open Optimize, you will see the
Optimize a model page. From here, you can create, edit or clone
Criteria Sets to apply to trained models, and view the results of evaluating a model
against a Criteria Set.
Tip: The model cannot be changed once the Criteria Set has been created. You
must create a new Criteria Set to use a different model.
To create a Criteria Set:
-
Click Create new Criteria Set.
-
Enter a Name for the model, and
Tags if required.
-
Select a trained model.
- This should be the model you want to apply these criteria to.
- The model cannot be changed once the Criteria Set has been created.
-
Click Save.
After saving a Criteria Set, the criteria definition page is displayed. Changes on
the criteria definition page are saved automatically.
You can come back to this page to edit an unoptimized Criteria Set by
right-clicking it on the Optimize a model page and clicking
Open.
By default
- all model input features are included in the Criteria Set unconstrained.
- output features are not assigned a target value and are equally weighted.
- Optional:
To apply a constraint to the model:
-
Select a constraint type in the Constraint column
(Full Range, Fixed or
Between).
-
For Fixed or Between
constraints, edit the allowed Value for the input
feature.
- Optional:
To set a target value for an optimization problem:
-
Set a goal type in the Goal column (Greater
than/Above, Less than/Below,
Between or None).
-
Enter a target Value for the output.
- The Value field auto-populates with the maximum
or minimum of the model Range.
- Click the Value field after selecting a goal type
of None to clear it.
-
Edit the Weight (importance) of the output relative to
the others.
-
To send a Criteria Set and its model to the engine, click
Optimize on the criteria definition page.
The card for your Optimization job will be displayed here and in the MI Job
Queue.