def load_dblp(data_ratio=0.9):
file_path = './datasets/mat/dblp-s.mat'
data = loadmat(file_path)
W = data['W'].astype(int)
label = data['gnd'].astype(int)
tot_len = W.shape[0]
# shuffle数组
idx = np.array(range(tot_len))
np.random.shuffle(idx)
W = W[idx, :]
W = W[:, idx]
label = label[idx]
data_len = np.int(W.shape[0] * data_ratio)
W = W[:data_len, :]
W = W[:, :data_len]
label = label[:data_len].reshape(-1)
# # 去除孤立点
degree = np.sum(W, axis=0).astype(int)
ZD = (degree.A != 0).reshape(-1)
W = W[ZD]
W = W[:, ZD]
label = label[ZD]
# 转化为one_hot 向量
label_set = set(label)
label = hd.one_hot(label, len(label_set))
return W, label
data_len = np.int(W.shape[0] * data_ratio)
W = W[:data_len, :]
W = W[:, :data_len]
label = label[:data_len].reshape(-1)
# # 去除孤立点
degree = np.sum(W, axis=0).astype(int)
ZD = (degree.A != 0).reshape(-1)
W = W[ZD]
W = W[:, ZD]
label = label[ZD]
# 转化为one_hot 向量
label_set = set(label)
label = hd.one_hot(label, len(label_set))
# 求关系矩阵
g1 = W
k = 2
for i in range(2, k, 1):
g1 = g1 * g1
W = W + g1
W = W / k
sm = scipy.sparse.csc_matrix(np.diag(np.sum(W, axis=0).A.reshape(-1) ** -.5))
NS = sm @ W @ sm
print("---初始化完成", '-' * 50)
# data = loadmat(file_path)
# print(data)
# network, labels = load_dblp(0.9)
tot_num = network.shape[0]
network += sparse.eye(tot_num)
D = sparse.csc_matrix(np.diag(np.sum(network, axis=0).A.reshape(-1)**-.5))
S = (D @ network @ D)
# attributes = sparse.eye(network.shape[0])
# network, labels = load_dblp(0.9)
# print(type(P[0]))
labels -= 1
label_set = set(labels)
label_num = len(label_set)
labels = hd.one_hot(labels, label_num)
train_ratio = 0.9
train_num = int(tot_num * train_ratio)
test_num = tot_num - train_num
temp = S
k = 5
P = np.empty((k, attributes.shape[0], attributes.shape[1]))
P[0] = S @ attributes
start = time.time()
for i in range(1, k, 1):
# temp = sparse.csr_matrix(temp @ S)
# P.append(sparse.csr_matrix(temp @ attributes))
temp = temp @ S
P[i] = temp @ attributes
print(time.time() - start)
