Dashboard widgets are the individual tiles that help create a dashboard view. Widgets provide an easy way to customize dashboards and perform ad hoc analysis. They can be used to build dashboards from scratch, or as a way to modify templated dashboards.
Widget Types
Click in each card to learn more about how to use each widget type
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag 'Timeseries' widget square
Define a the plot by specifying what metric you'd like to see, which Feature / Tag / Actual / Prediction Value you'd like to see first.
Overlay important metadata like tags by toggling on "Group metric by feature or tag"
Enjoy your powerful dashboard view!
🏁Track Key Slice Performance
How to create a widget within a dashboard that shows key slice performance over time
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag theTime Series widget creation button
Define the first plot by specifying the model, what metric you'd like to see, and the model environment: Production, Pre-production (Validation or Training) and version if applicable.
Duplicate the plot to quickly start defining your 2nd plot
Add a filter to specify the slice in the 2nd plot
Success!
Distribution
🧠Analyze Top Performing Features
How to create a widget within a dashboard that analyzes your top performing features
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag theDistribution widget creation button
Define the first plot by specifying the model, what metric you'd like to see, model environment: Production, Pre-production (Validation or Training), version (if applicable) and what (features, actuals, predictions, etc) you will be displaying the "distribution over".
💡 This example uses feature whose data type is numericdistribution. If you chose a feature whose data type is string, the values will be bucketed by that dimension's values instead of it's numeric ranges.
Success!
🔥Evaluate Performance Across Distributions with Heatmaps
How to create a widget within a dashboard where you can evaluate a heatmap of performance across distributions
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag theDistribution widget creation button
Define the base plot by specifying the model, what metric you'd like to see, model environment: Production, Pre-production (Validation or Training), version (if applicable) and what (features, actuals, predictions, etc) you will be displaying the "distribution over".
🔥Overlay performance information by selecting a performance metric in the Color By dropdown.
💡If the metric requires additional information like Positive Class or at K value, fill out those appropriate fields to get your finalized chart!
Success!
📊Compare Predictions vs Actuals
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag theDistribution widget creation button
Define the first plot by specifying the model, what metric you'd like to see, model environment: Production, Pre-production (Validation or Training), version (if applicable) and what (features, actuals, predictions, etc) you will be displaying the "distribution over". In this case, we're looking at Prediction Class
💾Duplicate the plot in the plot menu to quickly start defining your 2nd plot
Update the second plot to Actual Class to specify what the second plot will be distributing over
💡Further narrow down a plot by adding filters to specify problematic features within the main query
Success!
Statistic
🔑 Highlight Key Performance Metrics
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag 'Statistic' widget square
Define the widget by specifying the evaluation/performance metric. Then define the rest of the dataset by specifying model, what metric you'd like to see, model environment: Production, Pre-production (Validation or Training), version (if applicable).
💡To narrow down on this metric for a given features / actuals / predictions, etc add filters on the primary dataset
Success!
Drift
🔔 Identify Feature Changes Overtime
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag 'Drift' widget square
Select the model dimension to measure such as prediction/actual and feature/tag drift
❗Pro Tip: Gain a more granular view of how each slice impacts your metric by grouping your feature/tag
Success!
Alert Graph
📉 Visualize Sudden Model Changes
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag 'Alert Graph' widget square
Select a prediction drift and feature drift monitors for your model
❗Pro Tip: If you haven't created monitors yet, be sure to do so first! Learn more here.
Save, share, and troubleshoot by clicking the 'View Monitor' link.
❗Pro Tip: Learn how to troubleshoot drift monitors here.
🏥 General Model Health Check
Begin with an existing dashboard, blank dashboard, or start with a created template
Enter edit mode
Select or Drag 'Alert Graph' widget square
Select various model monitors that can significantly impact KPIs or are sensitive to change
❗Pro Tip: If you haven't created monitors yet, be sure to do so first! Learn more here.
Save, share, and troubleshoot by clicking the 'View Monitor' link.
❗Pro Tip: Learn how to troubleshoot various monitors here.
Text
✍️ Annotate Dashboard With Metadata
Begin with an existing dashboard, blank dashboard, or start with a created template
