Model Optimization

The Model Optimization module is used to define model-level parameter sets, evaluate tuning behavior in experiments, and promote stable configurations into production. In the current Lexmark UI, this module opens from Apps > Model Optimization and uses a card-style model list instead of the legacy tile flow. Each model then has tabbed work areas for Parameters, Experiments, Productions, and Settings.

This area is designed for iterative optimization: register a model, define error-type parameters, create experiments, and then maintain production-ready sets. The UI also supports direct metadata maintenance and delete flows from the settings tab. Availability of create/edit/delete actions depends on your role permissions.

Open Model Optimization

To open Model Optimization, do the following:

1. In the left sidebar, go to Apps and click Model Optimization.
2. Confirm the page header shows Model Optimization.
3. Confirm top-right actions include Create and Ask Optra (based on permissions).

Understand the Model List

The home page shows model cards with badges and relative update times so teams can quickly identify candidate models for tuning or maintenance. When no models exist, the page displays an onboarding empty state with guidance to create the first model.

1. Review model cards for name, badges, and Last updated metadata.
2. Click a model card to open its tabbed detail workspace.
3. If no models are listed, verify the No models found empty state and continue with model creation.

Register a New Model

To register a new model, do the following:

1. On the Model Optimization page, click Create.
2. In Register a new model, complete:
   • Name
   • Product Family
   • Rule Type
   • Description
3. Click Save.
4. The system opens the new model in the Parameters tab.

Use the Parameters Tab

The Parameters tab manages model error types and tuning bounds used by downstream experiment and production sets. The table supports search and row-level edit actions, and the top actions support both new parameter creation and parameter-copy workflows from another model.

1. Open a model and stay on the Parameters tab.
2. Use search to filter parameter rows.
3. Review parameter columns such as:
   • Error Type
   • Threshold (min, max, default)
   • Suppression Period (Days) (min, max, default)
   • Repetition Count Window (for repetition rule types)
   • Updated
4. Click a row to open Edit Error Type.
5. Use the trash action to open Remove Parameter and confirm deletion when needed.

Add a New Parameter (Add Error Type)

To add a new parameter, do the following:

1. In Parameters, click New Parameter.
2. In Add Error Type, click Select type and choose an error type.
3. Enter Description.
4. For supported model types, configure threshold/suppression/repetition min-max-default values.
5. Click Save.

Copy Parameters from Another Model

To copy parameters from another model, do the following:

1. In Parameters, click Copy Parameters.
2. In Copy Error types, choose a source model from Parameter set.
3. Review rows and confirm which types are importable (existing types are marked as already present).
4. Click Import.

Use the Experiments Tab

The Experiments tab is for non-production model tuning sets. It supports list search, quick drill-in, and creation of named experiment sets. This allows teams to iterate safely before promoting settings into production contexts.

1. Open the Experiments tab.
2. Use search to find experiment sets by name.
3. Review table columns (Name, Parameters, Updated).
4. Click New Experiment to open the creation modal.
5. Enter Experiment Name and Description, then click Create.
6. Click an experiment row to open experiment details/settings.

Configure and Run an Experiment

After creating an experiment set, add the error types to evaluate, define the historical analysis window, and run threshold optimization. Experiment runs are non-production: they do not change deployed sets or generate alerts.

1. Open an experiment set from the Experiments tab.
2. Click Add Parameter, then Select type and choose an error type.
3. Enter Suppression Period (Days) and Repetition Count Window. Threshold is not set here — it is evaluated across candidate values.
4. Set Input Date (inclusive end date of the window) and Past Period Data (months of history; defaults to 24 when blank).
5. Click Save, then repeat for each error type.
6. Click Run to start the optimization. Run is disabled while a run is in progress.
7. When the run completes, open the Performance tab to review results.

Review Experiment Performance Results

The Performance tab shows how candidate thresholds would have performed on historical data. Use it to compare thresholds and select a setting before promoting to production.

1. Open a completed experiment and click the Performance tab.
2. Review the four metric plots:
   • Precision (Valid Alert Rate)
   • Recall (Service Case Detection Coverage)
   • F-score (Balanced Alert Score)
   • Average Monthly Alert Volume
3. Review the Recommended Parameter Settings table for the recommended threshold per error type.
4. Select a threshold that fits the business priority (precision, recall, F-score, or alert volume).

Use the Productions Tab

The Productions tab tracks deployed sets that are active for operational use. This view mirrors experiments but focuses on production lifecycle and deployment context.

1. Open the Productions tab.
2. Use search to find production sets.
3. Review table columns (Name, Parameters, Updated) and deployment badges.
4. Click New Production.
5. In New Production, complete:
   • Production Name
   • Description
   • Parameter strategy (Copy parameters from experiment or create/select a new set, based on model type)
6. Click Create.
7. Click a production row to open production details/settings.

Configure and Deploy a Production Set

After creating a production set, add its error-type parameters and deploy it. The Error-Based Rule uses the deployed set on its next scheduled run. Creating or saving a set does not deploy it.

1. Open a production set from the Productions tab.
2. Click Add Parameter, then Select type and choose an error type.
3. Enter Threshold, Suppression Period (Days), and Repetition Count Window within the displayed ranges.
4. Click Save, then repeat for each error type.
5. Review the set, then click Deploy.
6. Confirm the set shows a Deployed badge in the Productions list.

Deployment activates the configuration; alerts are generated only when a later scheduled run finds data that meets the criteria.

Promote an Experiment to Production

Promotion creates a production set from an experiment, carrying over its error types and windows. Promotion does not deploy the set — apply the selected thresholds and deploy separately.

1. Open a completed experiment from the Experiments tab.
2. Click Promote to Production.
3. In the modal, review the Production Name (prefilled from the experiment), adjust if needed, then confirm with Yes, promote.
4. In the new production set, open each error type and enter the Threshold selected during the results review.
5. Click Save, then repeat for each error type.
6. Click Deploy, then confirm the Currently deployed badge.

Use Settings (Edit and Delete)

The Settings tab provides model metadata maintenance and irreversible delete actions. Use this area for title/description corrections and controlled retirement of models that are no longer needed.

1. Open the Settings tab.
2. Review model metadata and copyable Model ID.
3. Click Edit to open Edit model.
4. Update Title and/or Description, then click Save.
5. To remove the model, click Delete model.
6. In the confirmation modal, review the warning (This action can not be undone.) and click Delete only when intentional.

View Error-Based Rule Alerts

Alerts generated by a deployed Error-Based Rule appear in the Alert Queue under Alerts. Use this area to identify Error-Based Rule alerts, review details, and update status.

1. In the navigation sidebar, go to Operations and click Alerts.
2. Identify Error-Based Rule alerts by name.
3. Review list columns: Severity, Alert, Status, Last Alert, Device, and Account.
4. Click an alert to review its message, metadata, and timeline.
5. Set the status — Mark as open, Mark as case created, Resolve, or Dismiss — then save.
6. To view a device's full alert history, open the device under Devices and click on the Alerts tab.