Superset dashboards

After running drift queries in Superset SQL Lab, you can save results as datasets, create charts, and build monitoring dashboards for ongoing visibility into model drift.

Save query results as a dataset

To reuse a drift query result in charts and dashboards, save it as a dataset:

1. Run your drift query in SQL Lab.

2. From the Save drop-down, select Save dataset.

   

3. Enter a descriptive name for the dataset (e.g., tvd_drift_age_column or psi_categorical_drift).

4. Click Save.

tip

Use a consistent naming convention for datasets, such as <drift_method>_<feature_name> (e.g., tvd_drift_age, psi_drift_gender). This makes it easier to identify datasets when building dashboards.

Create a chart

To visualize drift data as a chart:

1. Go to Charts and click + Chart.

   

2. Select the dataset you saved.

3. Choose a chart type.

   

   Recommended chart types for drift monitoring:

   Chart type	Best for	Example use case
   Line chart	Drift score over time	Visualize how TVD or PSI changes across time buckets
   Bar chart	Comparing drift across columns	Side-by-side drift scores for multiple features
   Table	Detailed drift data	Show exact drift scores, sample sizes, and status labels
   Big Number with Trendline	Single-metric overview	Display the latest drift score with trend

4. Configure the chart:

   • Set the Time column to timestamp (or time_bucket).
   • Set the Metric to the drift score column (e.g., drift_score, psi_score, z_score_drift).
   • Optionally group by column_name to show multiple features on one chart.

5. Click Save, and add it to a new or existing dashboard.

Build a monitoring dashboard

To create a centralized view of your model's drift metrics:

1. Go to Dashboards and click + Dashboard.

   

2. Add your drift charts to the dashboard.

3. Arrange charts in a logical layout.

   

Recommended dashboard layout

Section	Charts to include
Overview	Big Number charts for latest drift scores per key feature
Numerical drift	Line charts showing TVD or Z-Score drift over time
Categorical drift	Line charts showing PSI drift over time
Detail tables	Table charts with exact scores, sample sizes, and drift status

Dashboard best practices

• Naming: Use descriptive dashboard names that include the model or deployment name (e.g., Credit Model - Drift Monitoring).
• Refresh schedule: Set an auto-refresh interval that matches your scoring frequency. For example, if you score hourly, set the dashboard to refresh every hour.
• Filters: Add dashboard-level filters for column_name to allow users to focus on specific features.
• Annotations: Add annotation layers to mark important events such as model retraining dates or data pipeline changes.
• Sharing: Use Superset's built-in sharing to distribute dashboard links to team members.

Dashboard visibility

Saved dashboards, charts, datasets, and saved queries are only visible to their creator by default. Other non-admin users will not see your dashboards in their dashboard list, nor will your charts appear in their chart list. To share with team members, add them as co-owners of the dashboard or use Superset's sharing feature. Admin users can see all objects across the platform.

Next steps

• Set up alerts to be notified when drift exceeds thresholds
• Export raw scoring data to Kafka for custom processing

---

Feedback

• Submit and view feedback for this page
• Send feedback about H2O MLOps to cloud-feedback@h2o.ai