5
$\begingroup$

I've just fit a binomial model (training y = 0 or 1) using R's gbm package. When I calculated predicted values using my validation data, some of the predicted values were less than 0.

Is this normal behavior? If so, how do I handle the negative values, floor them at 0?

$\endgroup$
5
$\begingroup$

I emailed the author of the package, Greg Ridgeway, and got this response:

"Predictions are on the canonical scale, which for binomial us the log-odds scale. You can use the type="response" when predicting to put it on the probability scale."

$\endgroup$

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.