Feature Analysis


These templates help you analyze the features a model uses as inputs for predictions. Feature-oriented dashboards enable quicker detection of issues, and are especially useful for deep troubleshooting to root cause problems. They can also be helpful aides in uncovering model retraining or improvement opportunities.

Feature Analysis Templates

Top Features
Feature Analysis
Feature Slice Performance
Feature Heatmap
This template allows you to look at any model features over any time period and environment (Training, Production and Validation).
  • Surfaces top features contributing to a model
  • Distribution view of each feature
  • Ability to look at conditional analysis through slices/filters
The feature analysis template is for feature analysis over time. The template lines up features based on the distribution and shows a day by day view.
  • Distribution view of feature
  • Day over day view of feature value
  • Ability to slice or filter based on input values
This template is designed to analyze the facets or slices of a single feature.
  • Designed for analysis of a single feature
  • It creates a set of bins or slices based on the distribution
  • It then shows performance of those bins based on an Evaluation Metric
  • the performance is shown over time for each slice
  • Numeric feature bins are created automatically
  • Categorical features, you do pick the values of your slices
The feature performance heat map breaks each feature into slices and color-coordinates them based on performance.
  • Overlay performance information on features
  • Show both volume and performance overlaid on slices of features
Questions? Email us at [email protected] or Slack us in the #arize-support channel