ott.core.gw_barycenter.GromovWassersteinBarycenter#

class ott.core.gw_barycenter.GromovWassersteinBarycenter(epsilon=None, min_iterations=5, max_iterations=50, threshold=0.001, jit=True, store_inner_errors=False, quad_solver=None, **kwargs)[source]#

Gromov-Wasserstein barycenter solver of the GWBarycenterProblem.

Parameters
  • epsilon (Optional[float]) – Entropy regulariser.

  • min_iterations (int) – Minimum number of iterations.

  • max_iterations (int) – Maximum number of outermost iterations.

  • threshold (float) – Convergence threshold.

  • jit (bool) – Whether to jit the iteration loop.

  • store_inner_errors (bool) – Whether to store the errors of the GW solver, as well as its linear solver, at each iteration for each measure.

  • quad_solver (Optional[GromovWasserstein]) – The GW solver.

  • kwargs (Any) – Keyword argument for GromovWasserstein. Only used when quad_solver = None.

Methods

init_state(problem, bar_size[, bar_init, a, ...])

Initialize the (fused) Gromov-Wasserstein barycenter state.

output_from_state(state)

No-op.

update_state(state, iteration, problem[, ...])

rtype

Tuple[float, bool, ndarray, Optional[ndarray]]

Attributes

is_low_rank

Whether the solver is low-rank.