This is continuation of the work started by @biyan-linkedin in PR #439. Thank you for your work Bill!
Previously, Photon would regularize the entire coefficient vector, including the intercept coefficient if one was present. One of our users had this to say:
Penalizing intercept makes the behavior of L2 regularization depends
on the pos/neg ratio, and often leads to hard-to-interpret intercept.
Especially when we have a lot of training data and most of the features
are extremely sparse.
This is continuation of the work started by @biyan-linkedin in PR #439. Thank you for your work Bill!
Previously, Photon would regularize the entire coefficient vector, including the intercept coefficient if one was present. One of our users had this to say: