Wasserstein disctance for multivariate distributions

[ignace-computing]

Dear package maintainers,

May I please ask a question?
Is there a way to get the following working?
In words: applying wasserstein for multivariate distributions.

using KernelFunctions
using ExactOptimalTransport
mu = discretemeasure(RowVecs(rand(7,3)))
nu = discretemeasure(RowVecs(rand(7,3)))
wasserstein(mu,nu)

This throws an error.

Many thanks!
Ignace

[zsteve]

Hi @ignace-computing ,
You seem to be wanting to solve discrete transport between multivariate empirical distributions, and it seems like wasserstein doesn't support that. I haven't got time to investigate in detail yet but my guess is that wasserstein only works for parametric distributions.
perhaps try out emd/emd2? these work for general discrete inputs and would require you to explicitly pass a cost matrix together with weights (in your case, uniform).