Hacker News new | past | comments | ask | show | jobs | submit login
The What-If Tool: Code-Free Probing of Machine Learning Models (2018) (googleblog.com)
134 points by polm23 12 days ago | hide | past | web | favorite | 7 comments





Everyone should check out the interpret-community Python package developed by Microsoft that encapsulates a lot of these what if scenarios. There's an equivalent "no-code" dashboard. More importantly, it relies heavily on SHAP where possible, which I think is the cat's pajamas.

I second this.

It is now in my default toolbox and almost all my models go through it.

I use it also to “teach“ data science.

Plus I met Rich Caruana who was presenting this at KDD, besides his awesome work (check out his paper on pneumonia), he’s a warm, simple and really receptive person.


Can you post a link?

Have a look at interpret.ml, which also links to their github org: https://github.com/interpretml

also here's the link to SHAP https://github.com/slundberg/shap

If you want to try out the what-if tool in a jupyter notebook (for free), create an account here on www.hopsworks.ai. It's our new SaaS ML platform, with a feature store. If you run the deep learning tour, it will populate a project with this notebook (it's open-source):

https://github.com/logicalclocks/hops-examples/blob/master/n...


Article is from 2018 fwiw.



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

Search: