[wills_forward]

Does anyone see explainability as another good reason to trees on tabular data, for which I think users would expect more digestable outputs?

[oofbey]

The kinds of trees that come out of these algorithms are so huge they really aren’t any more interpretable than a NN.

[CornCobs]

Not exactly. These tree models are ensemble methods, meaning they comprise several trees. Each individual tree may be small, but it is difficult to pinpoint explanations when that tree is but one amongst a forest

[aimor]

Yes, I've been looking at using decision trees for explaining models that are difficult to understand. Currently seeing useful results on real data sets. If you're interested, I've implemented parts of TREPAN [1] and it's very approachable. However it's also important to have interpretable features which is a whole other thing.

[1] https://research.cs.wisc.edu/machine-learning/shavlik-group/...