python实现标签匹配--匈牙利算法

# coding = utf-8
'''
用匈牙利算法实现标签的匹配问题，并且输出最后的标签
'''
import numpy as np
from scipy.optimize import linear_sum_assignment


label = np.array([1,1,2,1,1,2,2,2,3,2,2,3,1,3,3,2,3])
gnd = np.array([2,2,2,2,2,2,3,3,3,3,3,3,1,1,1,1,1])
K = np.unique(gnd)
K_num = len(K)

cost_mat = np.array(np.zeros((K_num,K_num)))

for i in range(K_num):
    temp_i = K[i]
    idx = np.where(np.array(label)==temp_i)
    for j in range(K_num):
        temp_j = K[j]
        h = gnd[idx]
        t = np.where(np.array(h)!=temp_j)
        cost_mat[i,j] = len(t[0])
print(cost_mat)

row_index,col_index = linear_sum_assignment(cost_mat)
assignment = col_index+1
print(assignment)
print('*'*200)
# assignment = assignment[::-1]
assignment_index = np.where(assignment)
result_label = np.array(np.zeros(label.shape))
for i in range(K_num):
    temp_i = K[i]
    print(temp_i)
    idx = np.where(np.array(label)==temp_i)
    print(idx)
    result_label[idx] = assignment[i]
print(result_label)
print(cost_mat[row_index,col_index])
print(cost_mat[row_index,col_index].sum())

参考链接

1. KMeans、LVQ、GaussianMixture几种聚类方法的Python实现以及标签映射（Kuhn-Munkres匈牙利算法）问题的解决(详细并附完整代码)
2. MATLAB实例：为匹配真实标签，对训练得到的标签进行调整