Row-normalize sparse matrix

import numpy as np
import scipy.sparse as sp
import torch

def normalize_row(mx):
    """Row-normalize sparse matrix"""
    rowsum = np.array(mx.sum(1))
    r_inv = np.power(rowsum, -1).flatten()
    r_inv[np.isinf(r_inv)] = 0.
    r_mat_inv = sp.diags(r_inv)
    mx = r_mat_inv.dot(mx)
    return mx.tocoo()

mx = sp.csr_matrix([[0,1,0,1],[0,0,0,1],[1,0,0,1],[0,1,0,0]])
a=normalize_row(mx)

rowsum = np.array(mx.sum(1)) #矩阵行求和  [[2],[1],[2],[1]]

运行到r_inv = np.power(rowsum, -1).flatten()出错了

ValueError: Integers to negative integer powers are not allowed.

rowsum的数据类型是整数，不是浮点数，只有浮点数才能进行power运算

于是我加了个小数点  mx = sp.csr_matrix([[0.0,1,0,1],[0,0,0,1],[1,0,0,1],[0,1,0,0]])

flatten()相当于flatten(0)

举个栗子 a的维度是（2，2），a这个数据从0维展开，就是（2 ∗ 2=4），维度就是(6)

a : tensor([[0.2,0.3],[0.5,0.6]])

a.flatten() : tensor([0.2,0.3,0.5,0.6])

r_inv[np.isinf(r_inv)] = 0.

如果r_inv存在inf（无穷）时赋值为0

  r_mat_inv = sp.diags(r_inv)就是用r_inv的值构造一个对角矩阵

.dot用法可以参考下：

.​​(4条消息) np.dot()使用方法_pillstap的博客-CSDN博客_np.dot

print(mx):

print(a):

print(a.A):

format

mx: csr_matrix

a:coo_matrix