Pre-configured dashboards enable users to quickly visualize, monitor and debug the health of the models and overall quality of data. As soon as the model is logged for monitoring, a set of default dashboards are created for you to get started quickly without any setup. The following dashboards are created by default:
Overview dashboard for a summary view
Performance dashboard with key model performance metrics
Drift dashboard that tracks feature and prediction drift values
Outlier dashboard to show outlier counts and statistical data quality checks
Create and manage dashboards
Dashboards in Verta help teams to create different views for different groups in order to get better visibility and faster troubleshooting of models. With customizable dashboards you can also create charts for production runbooks.
You can manage dashboards from the dashboard list view.
Add dashboard button will create a blank dashboard. You can then start adding panels and widgets.
Panels help create separate sections to help group similar metrics in one group. Use the add panel button to add a new section to your dashboard. You can edit to rename a panel and you also have the option to delete an existing panel.
A dashboard can have one more panels and each panel contains one or more widgets.
Widgets are different types of charts or tables that can be used to visualize different monitoring information.
Click the Create widget button to add a widget to a panel.
Select from a list of supported widgets to visualize your data
The ability to rename, edit and delete a widget is also available
Here is a list of supported widgets:
Single metric or summary stat (e.g. accuracy, prediction count)
Series of data points for one or more variables in the x-axis (e.g. accuracy metric, drift) against y-axis variable (e.g. time)
Plot a variable in the y-axis against another variable in the x-axis to depict patterns and correlation between the variables (e.g. outlier chart)
Plot and interpret the trade-off in performance for different threshold values (e.g. PR curve, ROC curve)
Summarized discrete or continuous data that are measured on an interval scale to depict distribution of data (e.g. distribution and drift charts)
Visualize data with rectangular bars (e.g. feature and prediction distribution charts)
A tabular view of features with statistics (e.g. drift table, outlier table)
Tabular description of the performance of a classification model using correct and incorrect predictions
Filters can help perform segmentation and cohort analysis. You can identify how model performance changes based on specific feature values and prediction values.