I 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

I 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...