We are currently using two versions of the prox: one which we have coded
manually (prox2) and one which is based on isotonic regression from sklearn.
Because of how things are coded, we need to use the manual one in the
implementation for NewtALM for numba to work.
We may as well just drop the other implementation since they are equally fast.
We are currently using two versions of the prox: one which we have coded manually (prox2) and one which is based on isotonic regression from sklearn. Because of how things are coded, we need to use the manual one in the implementation for NewtALM for numba to work.
We may as well just drop the other implementation since they are equally fast.