def compare_kmeans(C1, C2, unif):
"""
C1 and C2 are obtained from the attribute kmeans.labels_
"""
N = len(C1)
K, J = max(C1)+1, max(C2)+1
zeta = np.zeros((N,K))
gamma = np.zeros((N, J))
for i in range(N):
zeta[i, C1[i]] = 1
gamma[i, C2[i]] = 1
zeta = zeta.T
gamma = gamma.T
if unif:
alpha = [1./K for _ in range(K)]
beta = [1./J for _ in range(J)]
else:
alpha = [len(np.where(C1 == j)[0]) for j in range(K)]
beta = [len(np.where(C2 == j)[0]) for j in range(J)]
# compute EMD
d = lambda x,y: np.linalg.norm(x-y, ord=1)
emd = EMD(zip(zeta, alpha), zip(gamma, beta), d)
emd.write_lp_problem('emd_pb.lp')
emd.solve_lp()
return emd.distance
