Parameter setup for the *MNIST+USPS distance

[microsoft]

Parameter setup for the *MNIST+USPS distance

[chenmzh]

I ran the following following codebut get MNIST-USPS distance around 930 instead of 1260 mentioned in paper

# Load datasets
loaders_src = load_torchvision_data('MNIST',resize = 28,   maxsize=1000)[0]
loaders_tgt = load_torchvision_data('USPS', resize = 28,   maxsize=1000)[0]
dist = DatasetDistance(loaders_src['train'], loaders_tgt['train'],
                   inner_ot_method = 'exact',# gaussian_approx, exact, jdot and naive_upperbound
                   p = 2, entreg = 1e-2,inner_ot_entreg = 1e-2, 
                   device='cpu')
d = dist.distance(maxsamples = 10000)
print(f'OTDD(MNIST,USPS)={d}')

The gaussian_approx method gets an even smaller distance.
Thanks!