Getting insights with Watson OpenScale
You can track all the deployed models, also known as deployments, that you are monitoring through the IBM Watson OpenScale dashboard.
The Insights tab provides a high-level view of your deployment monitoring.
Deployments Monitored - In this example, a total of 25 deployments are being monitored. The deployments are shown as individual tiles. Ten of the 25 tiles are visible in the sample.
Quality Alerts - A total of seven Quality (previously called Accuracy) alerts are represented in the tiles and show values of various percentages.
Fairness Alerts - There are also seven Fairness alerts, which display in several of the tiles, represented by a small
BIAStag. The deployment shows various Fairness values as a percentage and also displays an indication of how many attributes are being reported.
Each tile provides a summary of monitoring activity for that deployment. Some deployment tiles show no bias yet, perhaps because fairness evaluation is not complete.
Fairness, quality, performance, and analytics insights
Select any of the individual deployment tiles to view more details about that deployment. Monitoring data for individual deployments displays in a series of charts. The charts track metrics, such as fairness, average requests per minute, and accuracy over days, weeks, or months.
- Viewing data for a deployment
- Visualizing data for a specific hour
- Debiasing options
Use the Explain a transaction tab to search a specific transaction ID to explain a particular transaction record.
- Explaining transactions
- Explaining categorical models
- Explaining image models
- Explaining unstructured text models
- Contrastive explanations
Model risk analysis and governance
Use the model risk and governance features to create pre-production and production repositories and compare models. For more information, see Model risk management and model governance.