If I were expressing a problem in terms of binary features, all encoded as {0,1}, could I boost some features by encoding them as {0,2}? Would the effect change based on whether I used either of the approaches above (logistic regression or linear SVMs?) Any paper recommendations that would explain the mathematics?
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
No, those are linear methods, the results would be equivalent. Multiplying $x$ by two would be offset by halfing the coefficient estimate.
answered Jun 29, 2019 at 14:42
-
$\begingroup$ i guess you’d need to tweak the optimization algorithm (e.g. gradient descent) to favor some of the parameters over others? $\endgroup$ Commented Jun 29, 2019 at 21:16