add getter for classProbabilities, or how do I visualise a decision tree? about mlpack

[flyingkukri]

Hi @rcurtin. Yes, that sounds good. Thaks a lot <3~! However I think it will be better if you open a pull request as I don't have any experience with open source contribution.

[rcurtin]

Would a getter for classProbabilities suffice? I spent a while thinking about it and it would be tricky to add a getter for the split value, since DecisionTree is not guaranteed to have a single split value. For instance, you could have an n-way split that split with (n - 1) different values into n children. Or, there may be no split value at all: in categorical dimensions, by default (with AllCategoricalSplit), the tree will split into all possible categorical values.

Anyway, the easiest thing to do here would be to provide ClassProbabilities() with documentation pointing out that the values are not class probabilities unless the node is a leaf. If that works for you, I'll open a PR (or you can open one, up to you). Alternately, if you have a proposal for how to provide a SplitValue() member in a safe and flexible way, I'm all ears. 😄

[rcurtin]

Sounds good, I opened #3511 to add the new feature. 👍

[rcurtin]

#3511 is merged, and so you can either use the git master branch, or wait for the next release of mlpack for the functionality to be stable. 👍