无监督K-Means算法（代码部分）

理论部分 https://jkchen.blog.csdn.net/article/details/103334497
这个比神经网络简单多了，一下午就KO了
顺便一提，可视化过程很有意思，也学到了很多matplotlib.pyplot的知识

代码：

import numpy as np
import matplotlib.pyplot as plt
from scipy.io import loadmat, savemat

X = loadmat('data_K_Means_2.mat')['X']
# X=np.random.rand(1000).reshape(500,2)
m, n = X.shape

# plt.scatter(X[:,0],X[:,1])
# plt.show()

times = 10  # 运行次数
color = ['red', 'yellow', 'black', 'pink', 'orange', 'purple']  # 每个簇的颜色
K = 6  # 簇数量
Show = True  # 是否可视化


# 两个数据集之间的距离平分
def Dis(i, j):
    return np.sum(np.multiply(i - j, i - j))


miJ = 1e18
ansU = []  # 簇中心
ansC = []  # 每个点的簇下标

index = np.random.permutation(m)
if Show:
    plt.ion()
for T in range(times):

    # 随机取样本点作为中心点
    subindex = np.random.choice(m, K)
    u = np.mat(np.zeros([K, n]))
    for i in range(K):
        u[i, :] = X[subindex[i], :]

    # 上一次的c
    cP = np.zeros([m])

    while True:
        if Show:
            plt.cla()

        c = np.zeros([m], int)  # 选择
        ct = np.zeros([K])  # 每个簇的点数
        sum = np.mat(np.zeros([K, n]))  # 求平均值

        # 找对应的c
        for i in range(m):
            mi = 1e18
            for j in range(K):
                dis = Dis(X[i], u[j])
                if dis < mi:
                    mi = dis
                    c[i] = j
            ct[c[i]] += 1
            sum[c[i]] += X[i]
            if Show:
                plt.plot(X[i, 0], X[i, 1], 'o', color=color[c[i]])

        # 找新的中心
        for i in range(K):
            if ct[i] > 0:
                u[i] = sum[i] / ct[i]

        # 画出中心
        if Show:
            for i in range(K):
                plt.plot(u[i, 0], u[i, 1], 'X', color='green')

        # 查看是否变化
        change = np.sum(np.abs(cP - c))
        cP = c
        if change == 0:
            J = 0  # 畸变函数
            for i in range(m):
                J += Dis(X[i], u[c[i]])
            if J < miJ:
                ansC = c
                ansU = u
                miJ = J

            if Show:
                name = 'Result_' + str(T) + '.png'
                plt.title('Finished! J is ' + str(J))
                print('save as :' + name)
                plt.savefig(name)

                plt.pause(1)
            else:
                print("Over!")
            break
        elif Show:
            plt.pause(0.01)

savemat('K_Means_Answer_2.mat', mdict={'U': ansU, 'C': ansC, 'J': miJ})

if Show:
    plt.ioff()
    for i in range(m):
        plt.plot(X[i, 0], X[i, 1], 'o', color=color[ansC[i]])
    for i in range(K):
        plt.plot(ansU[i, 0], ansU[i, 1], 'X', color='green')
    plt.title('Result : J is ' + str(miJ))
    name = 'Result Answer.png'
    plt.savefig(name)
    plt.show()

可视化效果：