Question

def pdist_numexpr(xs):
    a = xs[:, np.newaxis, :]
    return np.sqrt(ne.evaluate('sum((a-xs)**2, axis=2)'))

print pdist_numexpr(A)
%timeit pdist_numexpr(xs)

[[ 0.  5.]
 [ 5.  0.]]
10 loops, best of 3: 30.7 ms per loop