def next(self,input): # not preserve(net.initial_state)
t = self.net(input)
t.shape = (1,self.rsize)
self.last_state = t #saving state
self.net.initial_state = self.last_state #restore state
esnout = self.last_state.copy()
esnout = self.Out.process(esnout) #compute esnout
esnout = aux.normalize(esnout)
return esnout
def spectral_clustering_sym(S, k):
n, _ = S.shape
L = aux.laplacian(S)
D = np.diag(np.sum(S, axis=1))
D_half = np.sqrt(D)
D_minus_half = np.diag(1 / np.diag(D_half))
w, V = np.linalg.eigh(D_minus_half.dot(L).dot(D_minus_half))
index = np.argsort(w)
U = _stack_vec(V, k, index)
for row in U:
row = aux.normalize(row)
print(U)
_, label = cluster.vq.kmeans2(U, k)
return label.reshape((-1, 1))
def next(self,input, i = None): # not preserve(net.initial_state)
t = self.net(input)
t.shape = (1,self.rsize)
if i == None:
i = self.index.label(t)[0]
if (i >= len(self.stm)):
i = -1
self.last_state = t #saving state
self.net.initial_state = self.last_state #restore state
esnout = self.last_state.copy()
esnout = self.stm[i].process(esnout)
esnout = aux.normalize(esnout)
return esnout
