ott.problems.linear.potentials.DualPotentials

Contents

ott.problems.linear.potentials.DualPotentials#

class ott.problems.linear.potentials.DualPotentials(f, g, *, cost_fn, corr=False)[source]#

The Kantorovich dual potential functions \(f\) and \(g\).

\(f\) and \(g\) are a pair of functions, candidates for the dual OT Kantorovich problem, supposedly optimal for a given pair of measures.

Parameters:
  • f (Callable[[Array], float]) – The first dual potential function.

  • g (Callable[[Array], float]) – The second dual potential function.

  • cost_fn (CostFn) – The cost function used to solve the OT problem.

  • corr (bool) – Whether the duals solve the problem in distance form, or correlation form (as used for instance for ICNNs, see, e.g., top right of p.3 in [Makkuva et al., 2020])

Methods

distance(src, tgt)

Evaluate Wasserstein distance between samples using dual potentials.

plot_ot_map(source, target[, samples, ...])

Plot data and learned optimal transport map.

plot_potential([forward, quantile, ...])

Plot the potential.

transport(vec[, forward])

Transport vec according to Gangbo-McCann Brenier [Brenier, 1991].

Attributes

f

The first dual potential function.

g

The second dual potential function.