1

I'm trying to minimize the LASSO objective function using the Rcplex package in R. The help page of Rcplex shows that the objective function should be written in b'Qb + c'b form. How do I rewrite formula To something that has the form b'Qb +c'b or is there some different way to implement the LASSO objective function into Rcplex?

Muki
  • 31
  • 3
  • Are you sure Quadratic programming is the right algorithm for this? It would be more straightforward to implement this using gradient descent. – JAD Jun 21 '18 at 15:13
  • That would be a bit of an annoyance (e.g. l1 needs to be smoothed somehow). Any very good reason to this? In what context? As said: that's not a very good/specialized way of solving the problem; and if that is research / a prototype, you could start with the mostly automatic [cvx-based approach](https://cvxr.rbind.io/post/examples/cvxr_intro/), where all those things get done for you (at some cost; but we already established: you won't beat specialized algorithms). Some reformulations of this would also target SOCP instead of QP (expected to be a better way). – sascha Jun 23 '18 at 12:13

0 Answers0