3 ms·
Gig: A new method for explaining complex ensemble ML models
- ianandrich 7y agoInteresting. Has anyone played with this yet?
- manthideaal 7y agoI think that assigning credit to individual features doesn't work if there is a strong interaction among features. I would like to see a method of constructing a map that describes sets of features with strong interaction, that is new features that are not reducible to individual features in order to explain how the model work. Edited, (1) is a good first step at shapely value theory and applications: (1) https://medium.com/datalab-log/understanding-the-impact-of-features-and-data-through-shapley-values-f235489b0b3e https://medium.com/datalab-log/understanding-the-impact-of-f...
- cosmic_ape 7y agoNot a big fan of the idea of assigning credit to individual features, but as far as this idea goes, this seems like a very useful improvement. The blog mentions an evaluation on some loan related data, and this data does not appear in the paper. I wonder wether it is available somewhere.