python实现K均值聚类——K-means

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

N=1000     #随机生成的样本数
xlow=0    
xhigh=40
ylow=0
yhigh=40
zlow=0
zhigh=40
step=0      #记录聚类的次数

def randrange(n, vmin, vmax):
    r = np.random.rand(n)   #随机生成n 个介于0-1之间的数
    return (vmax-vmin) * r + vmin   #得到n个[vmin，vmax]之间的随机数

#准备数据
a = randrange(N, xlow, xhigh)
b = randrange(N, ylow, yhigh)
c = randrange(N, zlow, zhigh)

#计算欧氏距离
def calc_e_squire(x, y):
    dis = (x[0]- y[0]) ** 2 + (x[1] - y[1]) **2 + (x[2] - y[2]) **2
    return dis

m1=[]
m2=[]
m3=[]
m4=[]

for i in range(N):
    if (a[i] <= int((xlow+xhigh)/2))&(b[i] <= int((ylow+yhigh)/2)):
        m1.append(i)
    elif (a[i] <= int((xlow+xhigh)/2))&(b[i] >= int((ylow+yhigh)/2)):
        m2.append(i)
    elif (a[i] >= int((xlow+xhigh)/2))&(b[i] >= int((ylow+yhigh)/2)):
        m3.append(i)
    else:
        m4.append(i)

sjs1=random.randint(0, len(m1))
sjs2=random.randint(0, len(m2))
sjs3=random.randint(0, len(m3))
sjs4=random.randint(0, len(m4))
ste1=m1[sjs1]
ste2=m2[sjs2]
ste3=m3[sjs3]
ste4=m4[sjs4]

k1 = [a[ste1],b[ste1],c[ste1]]
k2 = [a[ste2],b[ste2],c[ste2]]
k3 = [a[ste3],b[ste3],c[ste3]]
k4 = [a[ste4],b[ste4],c[ste4]]

display1x=[]
display1y=[]
display1z=[]
display2x=[]
display2y=[]
display2z=[]
display3x=[]
display3y=[]
display3z=[]
display4x=[]
display4y=[]
display4z=[]

#定义聚类
while True:
    step+=1
    sse_k1 = []
    sse_k2 = []
    sse_k3 = []
    sse_k4 = []
    for i in range(N):
        e_squire1 = calc_e_squire(k1, [a[i], b[i], c[i]])
        e_squire2 = calc_e_squire(k2, [a[i], b[i], c[i]])
        e_squire3 = calc_e_squire(k3, [a[i], b[i], c[i]])
        e_squire4 = calc_e_squire(k4, [a[i], b[i], c[i]])
        
        if (e_squire1 <= e_squire2)&(e_squire1 <= e_squire3)&(e_squire1 <= e_squire4):
            sse_k1.append(i)
        elif (e_squire2 <= e_squire1)&(e_squire2 <= e_squire3)&(e_squire2 <= e_squire4):
            sse_k2.append(i)
        elif (e_squire3 <= e_squire1)&(e_squire3 <= e_squire2)&(e_squire3 <= e_squire4):
            sse_k3.append(i)
        else:
            sse_k4.append(i)
    
    
    k1_x = sum([a[i] for i in sse_k1]) / len(sse_k1)
    k1_y = sum([b[i] for i in sse_k1]) / len(sse_k1)
    k1_z = sum([c[i] for i in sse_k1]) / len(sse_k1)
    k2_x = sum([a[i] for i in sse_k2]) / len(sse_k2)
    k2_y = sum([b[i] for i in sse_k2]) / len(sse_k2)
    k2_z = sum([c[i] for i in sse_k2]) / len(sse_k2)
    k3_x = sum([a[i] for i in sse_k3]) / len(sse_k3)
    k3_y = sum([b[i] for i in sse_k3]) / len(sse_k3)
    k3_z = sum([c[i] for i in sse_k3]) / len(sse_k3)
    k4_x = sum([a[i] for i in sse_k4]) / len(sse_k4)
    k4_y = sum([b[i] for i in sse_k4]) / len(sse_k4)
    k4_z = sum([c[i] for i in sse_k4]) / len(sse_k4)
    
    if (k1 != [k1_x, k1_y, k1_z]) or(k2 != [k2_x, k2_y, k2_z]) or (k3 != [k3_x, k3_y, k3_z]) or (k4 != [k4_x, k4_y, k4_z]):
        k1 = [k1_x, k1_y, k1_z]
        k2 = [k2_x, k2_y, k2_z]
        k3 = [k3_x, k3_y, k3_z]
        k4 = [k4_x, k4_y, k4_z]
        display1x.append(k1_x)
        display1y.append(k1_y)
        display1z.append(k1_z)
        display2x.append(k2_x)
        display2y.append(k2_y)
        display2z.append(k2_z)
        display3x.append(k3_x)
        display3y.append(k3_y)
        display3z.append(k3_z)
        display4x.append(k4_x)
        display4y.append(k4_y)
        display4z.append(k4_z)
    else:
        break

label_font = {
'color' : 'k' ,
'size' : 25,
'weight' : 'bold'
}
    

fig = plt.figure(figsize=(16,12))   # 参数依然是图片大小
ax = fig.add_subplot(111, projection='3d')      #确定子坐标轴，
    
ax.set_xlabel("X axis", fontdict=label_font)
ax.set_ylabel("Y axis", fontdict=label_font)
ax.set_zlabel("Z axis", fontdict=label_font)
ax.set_title("K means clustering:", alpha=0.6, color="b", size=25, weight='bold')   #子图（其实就一个图）的title

    
kv1_x = [a[i] for i in sse_k1]
kv1_y = [b[i] for i in sse_k1]
kv1_z = [c[i] for i in sse_k1]
kv2_x = [a[i] for i in sse_k2]
kv2_y = [b[i] for i in sse_k2]
kv2_z = [c[i] for i in sse_k2]
kv3_x = [a[i] for i in sse_k3]
kv3_y = [b[i] for i in sse_k3]
kv3_z = [c[i] for i in sse_k3]
kv4_x = [a[i] for i in sse_k4]
kv4_y = [b[i] for i in sse_k4]
kv4_z = [c[i] for i in sse_k4]

q=0
w=0
r=0
t=0
sse_k1.append(0)
sse_k2.append(0)
sse_k3.append(0)
sse_k4.append(0)
distance=0 
for i in range(N):
    if sse_k1[q]==i:
        temp = calc_e_squire([a[i],b[i],c[i]],[k1_x,k1_y,k1_z])
        distance += math.sqrt(temp)/N
        q+=1
    elif sse_k2[w]==i:
        temp = calc_e_squire([a[i],b[i],c[i]],[k1_x,k1_y,k1_z])
        distance += math.sqrt(temp)/N
        w+=1
    elif sse_k3[r]==i:
        temp = calc_e_squire([a[i],b[i],c[i]],[k1_x,k1_y,k1_z])
        distance += math.sqrt(temp)/N
        r+=1
    elif sse_k4[t]==i:
        temp = calc_e_squire([a[i],b[i],c[i]],[k1_x,k1_y,k1_z])
        distance += math.sqrt(temp)/N
        t+=1

figure1 = ax.scatter(kv1_x,kv1_y,kv1_z,color='r') 
figure2 = ax.scatter(kv2_x,kv2_y,kv2_z,color='c') 
figure3 = ax.scatter(kv3_x,kv3_y,kv3_z,color='b') 
figure4 = ax.scatter(kv4_x,kv4_y,kv4_z,color='k') 

figure1 = ax.plot(display1x,display1y,display1z,color='k', linewidth=5)   
figure1 = ax.plot(display2x,display2y,display2z,color='k', linewidth=5)   
figure1 = ax.plot(display3x,display3y,display3z,color='k', linewidth=5)   
figure1 = ax.plot(display4x,display4y,display4z,color='k', linewidth=5)   

      
plt.show()
print ("----------聚类次数为" + str(step)+"次------")
print ("----------距离聚类中心点的平均距离为" + str(distance)+"------")