7 ms·
Does anyone see explainability as another good reason to trees on tabular data, for which I think users would expect more digestable outputs?
by wills_forward 4y ago
Does anyone see explainability as another good reason to trees on tabular data, for which I think users would expect more digestable outputs?
- oofbey 4y agoThe kinds of trees that come out of these algorithms are so huge they really aren’t any more interpretable than a NN.
- CornCobs 4y agoNot 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 4y agoYes, 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/craven.thesis.pdf https://research.cs.wisc.edu/machine-learning/shavlik-group/...