search

uber / manifold

A model-agnostic visual debugging tool for machine learning
Apache License 2.0
1.65k stars 117 forks source link

Not obvious distributions for categorical features #124

Open Lomdjaria1 opened 4 years ago

Lomdjaria1 commented 4 years ago

Not obvious distributions for categorical features

Summary

I want to see difference in Feature attribution window for two groups and for categorical feature, that has two possible values Mobile and Desktop. As I can See: Group0 has MOBILE: 1361 DESKTOP: 1478
Group1 has MOBILE: 3282 DESKTOP: 2823

Expected Behavior

I expected to see two bar charts, that reflects distribution differences

Current Behavior

I see barcharts that don't reflect anything

Screenshot 2020-04-10 at 14 28 04

Context and Environment

Browser - Chrome. npm 6.16, manifold v0.0.1 Behaviour is the same as in http://manifold.mlvis.io/

gnavvy commented 4 years ago

Thank you for bringing this up. @Lomdjaria1

The current design was optimized for features with multiple categories (ref). There's definitely space for improvement.

Mind opening a PR or suggest a design? : )

Lomdjaria1 commented 4 years ago

I don't understand how current layout was made? I suppose, that overlapped barcharts, with percentages of such type will be more informative. For example: I would like to see, that in group 1: mobile - is 12% precent of data, but in group1 - mobile is 40% . (on y_axis = 0.12 and 0.4)