How to compute Wasserstein distance?

64 views (last 30 days)
Arvind Ganesh
Arvind Ganesh on 23 May 2019
Edited: Georgios Pyrgiotakis on 14 May 2020
I have two probability distributions in arrays P and Q (normalized by area, of same length, and over same sample space). I want to find Wasserstein distance between them.
I tried to use the Earth Mover Distance from here:
My input was [Y, fval] = emd(X, X, P, Q, @gdf)
where X is the sample space. But it is giving output:
No feasible solution found.
Linprog stopped because no point satisfies the constraints.
Is this the way to compute Wasserstein distance? How do I fix this?

Answers (1)

Koorosh Aslansefat
Koorosh Aslansefat on 29 Apr 2020
Edited: Koorosh Aslansefat on 29 Apr 2020
I have recently added new set of functions for ECDF-based distance measure to MATLAB Exchange.
I would suggest to have a look.
  3 Comments
Koorosh Aslansefat
Koorosh Aslansefat on 13 May 2020
If P and Q can be considered as empirical distribution function (ECDF), then you can use those functions with a bit of modification.
Georgios Pyrgiotakis
Georgios Pyrgiotakis on 14 May 2020
Edited: Georgios Pyrgiotakis on 14 May 2020
Yes P and Q are emprical CDF. How can I use these functions? I tried but I get lost. It seems that you start with the data set convert it to CDF and then calculate the area between the two curves?

Sign in to comment.

Tags

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!