ott.tools.sinkhorn_divergence.SinkhornDivergenceOutput.to_dual_potentials

ott.tools.sinkhorn_divergence.SinkhornDivergenceOutput.to_dual_potentials#

SinkhornDivergenceOutput.to_dual_potentials(epsilon=None)[source]#

Return dual potential functions [Pooladian et al., 2022].

Using vectors stored in potentials, instantiate a dual potentials object that will provide approximations to optimal dual potential functions for the dual OT problem defined for the geometry stored in geoms[0]. These correspond to Equation 8 in [Pooladian et al., 2022].

Note

When static_b=True, the \(g\) potential function will not be debiased.

Parameters:

epsilon (Optional[float]) – Epsilon regularization. If None, use in geoms[0].

Return type:

DualPotentials

Returns:

The debiased dual potential functions.