The paper claims that the transition matrix T=D^0.5*E*D^0.5 is symmetric, and socialsent.graph_construct.transition_matrix takes a sym parameter that is supposed to make the transition matrix symmetric.
That is however not the case (mathematically, the above equation does not make a non-symmetric matrix symmetric). This can be trivially demonstrated by e.g.:
import numpy as np
from socialsent.graph_construction import transition_matrix
v = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]])
v_norm = np.divide(v.T, np.linalg.norm(v, axis=1)).T
E = namedtuple('Embedding', ['m'])
embeddings = E(m=v_norm)
M = transition_matrix(embeddings, nn=1, sym=True, arccos=True)
print M
The paper claims that the transition matrix
T=D^0.5*E*D^0.5
is symmetric, andsocialsent.graph_construct.transition_matrix
takes asym
parameter that is supposed to make the transition matrix symmetric.That is however not the case (mathematically, the above equation does not make a non-symmetric matrix symmetric). This can be trivially demonstrated by e.g.:
What am I missing?