Dashboard Widgets

Customize dashboards with widgets

Widget Overview

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

Time Series 📈

• Correlate a feature over time: compare to a feature/tag value
• Track key slice performance: a granular view of model performance

Distribution 📊

• Analyze top performing features with data quality metrics
• Evaluate the heat map: Performance across distributions
• Compare predictions vs actuals

Statistic ➕

• Highlight key performance metrics: for any model type, pair this with time series graphs for a single-pane-of-glass view of model health

Drift 🌊

• Identify gradual changes over time: measure drift for any model dimension

Alert Graph 🔔

• Visualize sudden model changes
• General check: combine business-critical monitors across various models in 1 dashboard

Text 📰

• Annotate Dashboards: add helpful notes and metadata to share across teams

Time Series

📈 Correlate a Feature Over Time

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 the Time 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 the Distribution 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 numeric distribution. 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 the Distribution 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 the Distribution 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

Enter edit mode

Select or Drag 'Text' widget square

Type useful notes and other relevant text

Enjoy your powerful dashboard view!

Questions? Email us at [e-mail] or Slack us in the #arize-support channel