mlfactor.github.io icon indicating copy to clipboard operation
mlfactor.github.io copied to clipboard

Published 20 hours ago • verified publisher icon shokru

Probably typo in Chapter 7 Xgboost

Open gelotran opened this issue 4 years ago • 4 comments

Hi,

Can you check the formula in the part 7.4.3 Xgboost? I think there may be a typo there.

image

Then the formula 7.5 has to be adjusted accordingly.

Best, Vu Tran

gelotran avatar Apr 17 '20 15:04 gelotran

Thanks for the feedback! I've checked again, and it seems ok to me, given the 2 factor in front of the sum...

shokru avatar Apr 17 '20 16:04 shokru

Hi,

Thank for feedback. I take a comparison with XGBoost website.

image

It seems that inside each tree leaf, we sum all the Hessian values but not a summation of \lamda.

Btw, your book is great for an R user working with Portfolio ML. I enjoy reading it a lot.

gelotran avatar Apr 17 '20 17:04 gelotran

True, but I prefer to take the simpler case when the loss is quadratic (easier to follow I find). And for a quadratic loss, the Hessian is constant. That's why I get differences w.r.t to the original derivation with "2" factors at some places. Thanks for the feedback again: don't hesitate because we really want to continuously improve the book.

shokru avatar Apr 17 '20 18:04 shokru

You're welcome. Thank for clarification

It's actually a good book. I hope that you can continue to work on it and looking to see further work from you.

gelotran avatar Apr 17 '20 19:04 gelotran