amueller / COMS4995-s20

COMS W4995 Applied Machine Learning - Spring 20
https://www.cs.columbia.edu/~amueller/comsw4995s20/
Creative Commons Zero v1.0 Universal
245 stars 114 forks source link

fix: binary cross entropy #3

Closed thomasjpfan closed 4 years ago

thomasjpfan commented 4 years ago

Without going into a derivation, one simple way to verifying that there is no negative is: -log(exp(z)+1)) is not convex.

log(exp(x)+1) is convex and the sum of convex functions are convex.

thomasjpfan commented 4 years ago

Or look at the scikit-learn docs on logistic regression