K均值聚类（无监督算法）

1.安装库

import numpy as np
import matplotlib.pyplot as plt
import random as rand
from mpl_toolkits.mplot3d import Axes3D
import random

2.定义函数

定义距离

定义初始聚类中心

附上标签

重新计算聚类中心

更新标签

检查标签是否需要再次更新

不断迭代

​def Euclidean_Distance(xi,xj):
    dist=np.sqrt(np.sum(np.square(xi-xj)))
    return dist

def init(K,X):#随机生成初始原型向量
    t = np.arange(0, K, 1)
    p=random.sample(list(A),K)#随机抽取k个点
    p=np.array(p)
    p1=np.ones_like(p)
    return p,t,p1

def mylabel(p, X, t):  # 定义距离标签
    print()
    sign = np.zeros((len(X[:, 0]),1))
    for i in range(len(X[:,0])):
        d = Euclidean_Distance(p[0], X[i])
        for j in range(len(p[:, 0])):
            temp = Euclidean_Distance(p[j], X[i])
            if temp <= d:
                d = temp
                sign[i] = t[j]
    return sign


def update(p,X,t):
    try0=mylabel(p,X,t)
    location_t=[]
    for i in range(len(t)):
        location_t.append(np.where(try0 == t[i])[0])
    p1=np.zeros_like(p)
    for i in range(len(t)):
        p1[i]=np.mean(X[location_t[i]],axis=0)
    plt.clf()# 清除当前 figure 的所有axes，但是不关闭这个 window，所以能继续复用于其他的 plot
    fig = plt.figure()
    ax = Axes3D(fig)
    ax.scatter(X[:, 0], X[:, 1], X[:, 2], c='grey')
    ax.scatter(Y[:, 0], Y[:, 1], Y[:, 2], c='white')
    ax.scatter(Z[:, 0], Z[:, 1], Z[:, 2], c='pink')
    plt.pause(2)
    return p1

def diff(p,p1):
    diff=np.mat(p-p1)
    diff_p=diff@diff.T
    e=np.arange(0,len(p[:,0]))
    for i in range(len(p[:,0])):
        e[i] = np.sqrt(diff_p[i,i])
    return e

def K_means(K,X):
    p,t,p1 = init(K,X)
    old_p=p1
    interval=0
    while max(diff(p,old_p))>0.0001:#设置聚类中心误差停止迭代的最大值
        old_p=p
        p=update(p,X,t)
        interval+=1
    return p

def random_norm(mu1, mu2, mu3, sigma1, sigma2, sigma3, r,size=1000):
   res = []
   chain = [mu1, mu2,mu3]
   for i in range(1000 + size):  # 这里的 1000 是为了让 Markov 链趋向极限分布
      # 计算给定 Y = chain[1] 时 X 的边际分布
      x = chain[0]
      chain[1] = rand.normalvariate(
          mu=mu2 + r * sigma2 / sigma1 * (x - mu1),
          sigma=sigma2 ** 2 * (1 - r ** 2))
      y = chain[1]
      chain[0] = rand.normalvariate(
         mu=mu1 + r * sigma1 / sigma2 * (y - mu2),
         sigma=sigma1 ** 2 * (1 - r ** 2))
      # 计算给定 X = chain[0] 时 Y 的边际分布
      z=chain[2]
      chain[2] = rand.normalvariate(
         mu=mu3,
         sigma=sigma3 ** 2)
      if i >= 1000:
         res.append(chain[:])
   return res
​

主函数

画出真实标签下的三类点

k均值分类的三类点

if __name__=='__main__':
    K=3
    t=np.arange(1,K+1,1)
    random.seed(45)
    X = np.array(random_norm(0, 0, 0, 2, 1, 4, 0.5, size=1000))
    Y = np.array(random_norm(20, 20, 20, 1, 2, 4, 0.5, size=1000))
    Z = np.array(random_norm(40, 40, 40, 1, 2, 2, 0.5, size=1000))
    A = np.vstack((X, Y, Z))
    y1 = np.zeros((X.shape[0], 1))
    y2 = np.ones((len(Y[:, 0]), 1))
    y3 = np.ones((len(Z[:, 0]), 1)) * 2
    label = np.vstack((y1, y2, y3))
    p,location_t=K_means(K,A)
    plt.clf()# 清除当前 figure 的所有axes，但是不关闭这个 window，所以能继续复用于其他的 plot
    ax = plt.figure().add_subplot(111, projection='3d')
    ax.scatter(X[:,0],X[:,1],X[:,2],c = 'w', marker = '^')
    ax.scatter(Y[:,0],Y[:,1],Y[:,2],c = 'pink', marker = '^')
    ax.scatter(Z[:,0],Z[:,1],Z[:,2],c = 'green', marker = '^')
    ax.scatter(p[:,0],p[:,1],p[:,2],c = 'b', marker = '^')
    ax.set_xlabel('X', fontdict={'size': 15, 'color': 'green'})
    ax.set_ylabel('Y', fontdict={'size': 15, 'color': 'green'})
    ax.set_zlabel('Z', fontdict={'size': 15, 'color': 'green'})
    fig = plt.figure()
    ax = Axes3D(fig)
    ax.scatter(A[location_t[0], 0], A[location_t[0], 1],A[location_t[0], 2], c='grey')
    ax.scatter(A[location_t[1], 0], A[location_t[1], 1],A[location_t[1], 2], c='white')
    ax.scatter(A[location_t[2], 0], A[location_t[2], 1],A[location_t[2], 2], c='pink')
    plt.show()