RuleMatrix: Visualizing and Understanding Classifiers using Rules

Yao Ming, Huamin Qu, and Enrico Bertini


Summary

RuleMatrix is an interactive visualization technique that helps users understand the input-output behavior of machine learning models. By viewing a classifier as a black box, we extract a standardized rule-based knowledge representation (a rule list) from its input-output behavior. The extracted rule list is presented as RuleMatrix, a matrix-based visualization, to help users navigate and verify the rules and the black-box model.

Software

The software RuleMatrix and its documentation can be found on Github.

Video Tutorial

Publication

Yao Ming, Huamin Qu, Enrico Bertini. RuleMatrix: Visualizing and Understanding Classifiers using Rules. IEEE Transactions on Visualization and Computer Graphic, 2018 (to appear).

Materials

[preprint] | [code]