This module implements various solvers to estimate optimal transport between two probability measures, through samples, parameterized as neural networks. These neural networks are described in ott.neural.models, borrowing lower-level components from ott.neural.layers using flax.


map_estimator.MapEstimator(dim_data, model)

Mapping estimator between probability measures.

neuraldual.W2NeuralDual(dim_data[, ...])

Solver for the Wasserstein-2 Kantorovich dual between Euclidean spaces.

neuraldual.BaseW2NeuralDual([parent, name])

Base class for the neural solver models.

Conjugate Solvers#

conjugate.FenchelConjugateLBFGS([gtol, ...])

Solve for the conjugate using LBFGS.


Abstract conjugate solver class.

conjugate.ConjugateResults(val, grad, num_iter)

Holds the results of numerically conjugating a function.