|
Dear all, I try to use logistic regression + L1 regularization, for the general formulation of LR+L1. In liblinear, it has supplied an standard LR+L1 solution, but it is an unconstraint solution, where the weight vector W is arbitrary. If I wish W to be larger than zero as W>0, for this constraint condition, what shall I do to get my solution? Thank you! |
|
If you are not using liblinear but your own solver you can add positivity constraint as additional regularizer and implement it as, for example, as Split-Bregman. There are number of paper on combination of L1 and positivity constraint as Split-Bregman, I have used it myself and it was working. I'm not working as liblinear, so don't know if it allow plug-in a code for new regularizer. |
