FreeGrad

• Paper
• Arxiv

Freegrad is a base learning algorithm contributed by @zmhammedi. It is described in the associated paper. It accepts simple labels and produces scalar predictions.

Options

FreeGrad Options:
  --freegrad            Diagonal FreeGrad Algorithm
  --restart             Use the FreeRange restarts
  --project             Project the outputs to adapt to both the lipschitz and
                        comparator norm
  --radius arg          Radius of the l2-ball for the projection. If not
                        supplied, an adaptive radius will be used.
  --fepsilon arg (=1, ) Initial wealth