It works very well for some problems but not for others. See some results below. The sparse implementation is for instance slow and sometimes the inner solver just won't converge (Scheetz2006 for instance).
Much of the work here was done by @JonasWallin .
[^1]: Z. Luo, D. Sun, K.-C. Toh, and N. Xiu, “Solving the OSCAR and SLOPE models using a semismooth Newton-based augmented Lagrangian method,” Journal of Machine Learning Research, vol. 20, no. 106, pp. 1–25, 2019.
This PR adds support for the Newt-ALM solver[^1].
It works very well for some problems but not for others. See some results below. The sparse implementation is for instance slow and sometimes the inner solver just won't converge (Scheetz2006 for instance).
Much of the work here was done by @JonasWallin .
[^1]: Z. Luo, D. Sun, K.-C. Toh, and N. Xiu, “Solving the OSCAR and SLOPE models using a semismooth Newton-based augmented Lagrangian method,” Journal of Machine Learning Research, vol. 20, no. 106, pp. 1–25, 2019.