Ryan Curtin

**Commented by rcurtin on 27 Jul 41689222 11:19 UTC** I have been investigating how to do this and I actually think it may be possible to do this entirely in...

**Commented by rcurtin on 8 Nov 41699463 22:40 UTC** (In #21) No more blocked by #20; let's call that it's own ticket to add profiling support, not to replace what...

The issue has not been closed so it is not done. You are free to look into it.

Sorry for the slow response. This actually is not yet done, but, note that initiatives like #2646 and #2440 mean that the dependencies of mlpack will change significantly in the...

Yeah, that's fine---in either case, when instance weights are added, it should be very easy to generalize the MDL support.

This sounds reasonable to me. We haven't used any mock objects for testing but I am not sure that there is any particular reason. Personally I think it's adequate to...

That seems reasonable, but consider adding the penalty as, e.g., a fourth template template parameter to the `DecisionTree` class. The default could be `NoPenalty` or something like that, and then...

Yeah, that seems like a reasonable approach to me.

Sorry, I hadn't gotten to this yet. If no split is found, then in my view there's no reason to add an MDL penalty because we wouldn't be adding any...

Ah, right, sorry that I misunderstood your question. Thanks for the clarification. Here is the paper that introduced the MDL penalty: https://arxiv.org/pdf/cs/9603103.pdf The relevant part is on page 4 in...