ott.tools.progot.ProgOT

Contents

ott.tools.progot.ProgOT#

class ott.tools.progot.ProgOT(alphas, *, epsilons=None, epsilon_scales=None, is_debiased=False)[source]#

Progressive Entropic Optimal Transport solver [Kassraie et al., 2024].

Parameters:
  • alphas (Array) – Stepsize schedule of shape [num_steps,].

  • epsilons (Optional[Array]) – Epsilon regularization schedule of shape [num_steps,]. If None, use the default epsilon at each step.

  • epsilon_scales (Optional[Array]) – Scale for the default epsilon of shape [num_steps,]. If None, don’t scale the epsilons. Note that only one of epsilons and epsilon_scales can be passed.

  • is_debiased (bool) – Whether to use sinkhorn_divergence() or Sinkhorn.

Methods