A few short demos illustrating some properties of the Wasserstein-2 distance.
python wass.py desired_resolution
where desired_resolution tells you how many transformations occur.
This test is done through do_shift_test()
This test is done through do_amp_test(). NOTE: This property is also held by the L^2 norm, and the current status of the code shows that W2 is
quite sensitive to this. This could be caused by my choice of renormalization operator. If one modifies this to a Gaussian mixture, I believe it would
work, since then renormalization is not necessary or simply is dividing by a number, rather than splitting into positive and negative parts
(a non-smooth operation).
This test is done through do_noise_test().