🔥 LIT v0.5: Learning Interpretability Tool |
This is a major release, including new modules (Tabular Feature Attribution and Dive) and many UI, dataset, and performance improvements.
LIT also has a new name, the Learning Interpretability Tool, to reflect interpretability features for image and tabular data in addition to text.
Check the release notes for more. |
|
Tabular Feature Attribution Find the most influential features affecting model predictions by computing a SHAP value for each feature. Explore polarity and feature influence in a heatmap table. See a tutorial on how to analyze feature importance in the Penguins demo.
|
|
DiveDive provides an interactive interface that groups data by feature value, with each square representing a datapoint. Easily spot patterns and outliers in complex data sets, to help identify systematic errors or evaluate ground truth. Try it out in the Penguins demo or find more information in documentation.
|
Dataset ImprovementsBetter performance on large datasets (up to 100k examples). Export data with CSV download, or access selection directly from Python in notebook mode to more easily integrate with other tools.
|
Please reach out to us through GitHub issues if you have any questions.
Want to subscribe to our updates? Join the lit-announcements group.
|
Made with 🔥 by the LIT team |
