0
$\begingroup$

Usually my job is to do classification but recently I have a project which requires me to do regression. That is to say, my response variable is not a binary True/False, but a continuous number. the response is a well formed Gaussian centered at zero.

I usually use XGboost for my classification tasks. To do a regression, I found the following from xgboost's manual:


General parameter

this is the only difference with classification, use reg:linear to do linear regression

when labels are in [0,1] we can also use reg:logistic

objective = reg:linear


I guess it means: regression, using a linear model

However, I looked at the output and I found that the output is still a set of trees. As we know, tree is no linear. So what does reg:linear really mean? From my perspective, there isn't any "linear regression" here.

Can anyone provide any insight here?

Thanks

$\endgroup$
5
$\begingroup$

A regression tree makes sense. You 'classify' your data into one of a finite number of values. Note, that while called a regression, a regression tree is a nonlinear model. Once you believe that, the idea of using a random forest instead of a single tree makes sense. One just averages the values of all the regression trees. Once one has a regression forest, the jump to a regression via boosting is another small logical jump. You are just running a bunch of weak learner regression trees and re-weighting the data after each new weak learner is added into the mix. So boosting can give a 'regression', but it a very non-linear model ! The details are explained in more detail and in more clarity here - https://homes.cs.washington.edu/~tqchen/pdf/BoostedTree.pdf . I believe he is one of the authors of the xgboost package.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.