moscot.backends.ott.GraphOTTOutput#

class moscot.backends.ott.GraphOTTOutput(output, shape)[source]#

Output of OT problems with a graph geometry in the linear term.

Parameters:

output (Union[SinkhornOutput, LRSinkhornOutput, GWOutput, LRGWOutput]) – Output of the ott backend.
shape (Tuple[int, int]) – Shape of the problem.

Methods

`as_linear_operator`([scale_by_marginals])	Transform `transport_matrix` into a linear operator.
`chain`(outputs[, scale_by_marginals])	Chain subsequent applications of `transport_matrix`.
`plot_costs`([last, title, return_fig, ax, ...])	Plot regularized OT costs during the iterations.
`plot_errors`([last, title, outer_iteration, ...])	Plot errors along iterations.
`pull`(x[, scale_by_marginals])	Pull mass through the `transport_matrix`.
`push`(x[, scale_by_marginals])	Push mass through the `transport_matrix`.
`sparsify`(mode[, value, batch_size, ...])	Sparsify the `transport_matrix`.
`to`([device])	Transfer self to another compute device.

Attributes

`a`	Marginals of the source distribution.
`b`	Marginals of the target distribution.
`converged`	Whether the solver converged.
`cost`	Regularized OT cost.
`dtype`	Underlying data type.
`is_linear`	Whether the output is a solution to a linear problem.
`is_low_rank`	Whether the `transport_matrix` is low-rank.
`potentials`	Dual potentials \(f\) and \(g\).
`rank`	Rank of the `transport_matrix`.
`shape`	Shape of the `transport_matrix`.
`transport_matrix`	Transport matrix of shape `[n, m]`.