moscot.backends.ott.GWSolver#

class moscot.backends.ott.GWSolver(jit=True, rank=-1, initializer=None, initializer_kwargs=mappingproxy({}), linear_solver_kwargs=mappingproxy({}), **kwargs)[source]#

Solver for the quadratic problem [Mémoli, 2011].

The Gromov-Wasserstein (GW) problem involves two distribution in possibly two different spaces. Points in the source distribution are matched to points in the target distribution by comparing the relative location of the points within each distribution.

Parameters:

jit (bool) – Whether to jit() the solver.
rank (int) – Rank of the solver. If -1 use the full-rank GW [Peyré et al., 2016], otherwise, use the low-rank approach [Scetbon et al., 2021].
initializer (Optional[Literal['random', 'rank2', 'k-means', 'generalized-k-means']]) – Initializer for GromovWasserstein.
initializer_kwargs (Mapping[str, Any]) – Keyword arguments for the initializer.
linear_solver_kwargs (Mapping[str, Any]) – Keyword arguments for Sinkhorn or LRSinkhorn, depending on the rank.
kwargs (Any) – Keyword arguments for GromovWasserstein .

Methods

__call__([xy, x, y, tags, device])

Solve an optimal transport problem.

Attributes

`is_fused`	Whether the solver is fused.
`is_low_rank`	Whether the `solver` is low-rank.
`problem_kind`	Problem kind this solver handles.
`rank`	Rank of the `solver`.
`solver`	`ott` solver.
`x`	The first geometry defining the quadratic term.
`xy`	Geometry defining the linear term in the FGW.
`y`	The second geometry defining the quadratic term.