def wavelet_origin(dataset,
order,
threshold,
s,
mask,
data_normalize,
adj,
laplacian_normalize,
sparse_ness=False):
from weighting_func import laplacian, fourier, weight_wavelet
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(dataset, L)
Weight = weight_wavelet(s, lamb, U)
del lamb, U, L
if (sparse_ness):
# for i in range(Weight.data.shape[0]):
# if(Weight.data[i] < threshold):
# Weight.data[i] = 0.0
Weight[Weight < threshold] = 0
Weight = sp.csr_matrix(Weight)
Weight.setdiag([0.0] * Weight.shape[0])
if (data_normalize == True):
Weight = normalize(Weight, norm='l1', axis=1)
t_k = [Weight]
t_k.append(sp.eye(adj.shape[0]))
return sparse_to_tuple(t_k)
def wave_gcn(order,
threshold,
s,
mask,
data_normalize,
weight_normalize,
adj,
laplacian_normalize,
sparse_ness=False):
from weighting_func import laplacian, fourier, weight_wavelet
# wavelet community
# adj = adj_matrix()
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(L)
# print np.dot(np.dot(U,np.diag(lamb)),np.transpose(U))
# U 列 eigen function
Weight = weight_wavelet(s, lamb, U)
t_k = to_gcn_weighting_function(order,
threshold,
mask,
data_normalize,
weight_normalize,
adj,
Weight,
sparse_ness=sparse_ness)
# return U,Weight, t_k
return sparse_to_tuple(t_k)
def spectral_basis(dataset, adj, s, laplacian_normalize, sparse_ness,
threshold, weight_normalize):
from weighting_func import laplacian, fourier, weight_wavelet, weight_wavelet_inverse
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(dataset, L)
U = sp.csr_matrix(U)
# U_transpose = sp.csr_matrix(np.transpose(U))
t_k = [U]
return sparse_to_tuple(t_k)
def wavelet_basis_chebyshev(dataset, adj, s, laplacian_normalize, sparse_ness,
threshold, weight_normalize):
from weighting_func import laplacian, fourier, weight_wavelet, weight_wavelet_inverse
from pygsp import graphs, filters
G = graphs.Graph(adj)
taus = [s]
g = filters.Heat(G, taus)
signal_matrix = np.identity(G.N)
Weight = g.filter(signal_matrix, method='chebyshev', order=50)
# taus = [-s]
# g = filters.Heat(G, taus)
# signal_matrix = np.identity(G.N)
# inverse_Weight = g.filter(signal_matrix, method='chebyshev', order=30)
# 逆变换和正变换s不能共享，否则不可逆
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(dataset, L)
# Weight = weight_wavelet(s, lamb, U)
inverse_Weight = weight_wavelet_inverse(1.0, lamb, U)
del U, lamb
if (sparse_ness):
Weight[Weight < threshold] = 0.0
inverse_Weight[inverse_Weight < threshold] = 0.0
print len(np.nonzero(Weight)[0])
print len(np.nonzero(inverse_Weight)[0])
if (weight_normalize == True):
Weight = normalize(Weight, norm='l1', axis=1)
inverse_Weight = normalize(inverse_Weight, norm='l1', axis=1)
Weight = sp.csr_matrix(Weight)
inverse_Weight = sp.csr_matrix(inverse_Weight)
print Weight, inverse_Weight
t_k = [inverse_Weight, Weight]
# t_k = [Weight]
return sparse_to_tuple(t_k)
def wavelet_basis(dataset, adj, s, laplacian_normalize, sparse_ness, threshold,
weight_normalize):
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(dataset, L)
Weight = weight_wavelet(s, lamb, U)
inverse_Weight = weight_wavelet_inverse(s, lamb, U)
del U, lamb
if (sparse_ness):
Weight[Weight < threshold] = 0.0
inverse_Weight[inverse_Weight < threshold] = 0.0
# print len(np.nonzero(Weight)[0])
if (weight_normalize == True):
Weight = normalize(Weight, norm='l1', axis=1)
inverse_Weight = normalize(inverse_Weight, norm='l1', axis=1)
Weight = sp.csr_matrix(Weight)
inverse_Weight = sp.csr_matrix(inverse_Weight)
# print Weight
t_k = [inverse_Weight, Weight]
return sparse_to_tuple(t_k)
def wave_basis(dataset, adj, s, laplacian_normalize, sparse_ness, threshold,
weight_normalize):
from weighting_func import laplacian, fourier, weight_wavelet, weight_wavelet_inverse
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(dataset, L)
Weight = weight_wavelet(s, lamb, U)
inverse_Weight = weight_wavelet_inverse(s, lamb, U)
del U, lamb
# positive_num = Weight[Weight > 0.0].shape[0]
# nonzero_num = len(np.nonzero(Weight)[0])
# print positive_num,nonzero_num
# positive_num = inverse_Weight[inverse_Weight > 0.0].shape[0]
# nonzero_num = len(np.nonzero(inverse_Weight)[0])
# print positive_num, nonzero_num
# inverse_Weight[inverse_Weight > -threshold] = 0.0
# print len(np.nonzero(inverse_Weight)[0])
# Weight[Weight > -threshold] = 0.0
# print len(np.nonzero(Weight)[0])
if (sparse_ness):
Weight[Weight < threshold] = 0.0
inverse_Weight[inverse_Weight < threshold] = 0.0
print len(np.nonzero(Weight)[0])
print len(np.nonzero(inverse_Weight)[0])
if (weight_normalize == True):
Weight = normalize(Weight, norm='l1', axis=1)
inverse_Weight = normalize(inverse_Weight, norm='l1', axis=1)
Weight = sp.csr_matrix(Weight)
inverse_Weight = sp.csr_matrix(inverse_Weight)
print Weight
t_k = [inverse_Weight, Weight]
# t_k = [Weight,inverse_Weight]
# t_k = [Weight]
return sparse_to_tuple(t_k)
def wavelet_gcn_origin(order,
threshold,
s,
mask,
data_normalize,
adj,
laplacian_normalize,
sparse_ness=False):
from weighting_func import laplacian, fourier, weight_wavelet
L = laplacian(adj, normalized=laplacian_normalize)
lamb, U = fourier(L)
print U.shape, lamb[0], lamb[1]
# print np.dot(np.dot(U,np.diag(lamb)),np.transpose(U))
# U 列 eigen function
Weight = weight_wavelet(s, lamb, U)
if (mask == True):
# mask 需要保留自身，为了使得自身的影响和其他可比，对adj加单位阵
if (order == 0):
adj = sp.csr_matrix(np.eye(adj.shape[0]))
elif (order == 1):
adj = adj + sp.csr_matrix(np.eye(adj.shape[0]))
elif (order == 2):
second_order = adj.dot(adj)
adj = adj + sp.csr_matrix(np.eye(adj.shape[0])) + second_order
del second_order
elif (order == 3):
second_order = adj.dot(adj)
third_order = adj.dot(second_order)
adj = adj + sp.csr_matrix(np.eye(
adj.shape[0])) + second_order + third_order
del second_order, third_order
elif (order == 4):
second_order = adj.dot(adj)
third_order = adj.dot(second_order)
four_order = adj.dot(third_order)
adj = adj + sp.csr_matrix(np.eye(
adj.shape[0])) + second_order + third_order + four_order
del second_order, third_order, four_order
Weight = adj.multiply(Weight)
if (sparse_ness):
# for i in range(Weight.data.shape[0]):
# if(Weight.data[i] < threshold):
# Weight.data[i] = 0.0
Weight[Weight < threshold] = 0
# 一阶关系利用GCN拟合，其他关系利用wavelet
print len(np.nonzero(Weight)[0])
Weight = np.multiply(Weight, (np.ones(shape=adj.shape) - adj))
# Weight = Weight * (np.ones(shape=adj.shape) - adj)
print len(np.nonzero(Weight)[0])
Weight = sp.csr_matrix(Weight)
Weight.setdiag([0.0] * Weight.shape[0])
print len(np.nonzero(Weight)[0])
if (data_normalize == True):
Weight = normalize(Weight, norm='l1', axis=1)
print Weight.shape
print len(np.nonzero(Weight)[0])
print Weight
# print Weight.todense()
# Weight = normalize(Weight + sp.eye(adj.shape[0]),norm='l1',axis = 1)
t_k = [Weight]
# 一阶和自身关系利用GCN拟合
t_k.append(preprocess_adj(adj))
# t_k.append(sp.eye(adj.shape[0]))
return sparse_to_tuple(t_k)
