Measure feature importance as the KL divergence between model outputs w/ feature randomized and original features: gives importance in bits!
related thing we've done is to randomly pick a side for the feature we're "permuting" at each split, based on weight of subtree
-
-
... not sure that made sense. Can elaborate later :)
-
I think I see where you're going there. I've never done it, but sounds a bit like an idea I had in mind.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.