Hacker News new | past | comments | ask | show | jobs | submit login
Manifold: A model-agnostic visual debugging tool for machine learning (2019) (uber.com)
68 points by pplonski86 on Feb 8, 2020 | hide | past | favorite | 5 comments



Reading that whole post just looking forward to the github link.

I was sorely disappointed ️



One thing that seems to be missing is a way to straightforwardly look at individual examples of data points in each cluster. Oftentimes looking at a few images (if that's what you're classifying) helps a lot more than staring at feature distribution plots.


It was exactly what I thought. Especially they explicitly say

>"With Manifold’s design, we turned the traditional ML model visualization challenge on its head. Instead of inspecting models, we inspect individual data points "

It really seems that they went on implanting the more sophisticated step (which seems very promising), while skipping the straight forward one.

What am I missing?


Sounds like what What-It Tool (https://pair-code.github.io/what-if-tool/) provides.




Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: