1
$\begingroup$

Are any kind of normalizations useful when using mutual information to select features?

The Wikipedia page has a few variants but they don't seem helpful. The entropy of the response is a constant and the entropy of the features ($H(X)$) doesn't seem relevant since they are fixed when I'm making predictions.

Improve this question
$\endgroup$
2
  • $\begingroup$ It seems this question has been answered here, stats.stackexchange.com/questions/1357/… $\endgroup$
    – Nikolas Rieble
    Commented Aug 4, 2018 at 5:42
  • $\begingroup$ Normalizing is not just scaling since $H(X)$ depends on which feature is being evaluated. $\endgroup$
    – badmax
    Commented Aug 7, 2018 at 14:11

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.