num=10 #k的種類
for k in range(1,num):
 隨機(jī)選擇k個(gè)質(zhì)點(diǎn)
 for i in range(n): #迭代n次
 根據(jù)點(diǎn)與質(zhì)點(diǎn)間的距離對(duì)于X_train進(jìn)行聚類
 根據(jù)mean值迭代更新質(zhì)點(diǎn)
 計(jì)算SSE_train
 計(jì)算SSE_test
畫圖

import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
from sklearn.model_selection import train_test_split 

def distance(p1,p2):
 return np.sqrt((p1[0]-p2[0])**2+(p1[1]-p2[1])**2)

def means(arr):
 return np.array([np.mean([p[0] for p in arr]),np.mean([p[1] for p in arr])])

#數(shù)據(jù)處理
data= pd.read_table('cluster.dat',sep='\t',header=None) 
data.columns=['x']
data['y']=None
for i in range(len(data)): #遍歷每一行 
 column = data['x'][i].split( ) #分開第i行，x列的數(shù)據(jù)。split()默認(rèn)是以空格等符號(hào)來分割，返回一個(gè)列表 
 data['x'][i]=column[0] #分割形成的列表第一個(gè)數(shù)據(jù)給x列 
 data['y'][i]=column[1] #分割形成的列表第二個(gè)數(shù)據(jù)給y列
list=[]
list1=[]
for i in range(len(data)):
 list.append(float(data['x'][i]))
 list.append(float(data['y'][i]))
 list1.append(list)
 list=[]
arr=np.array(list1)
print(arr)

#按照8：2劃分?jǐn)?shù)據(jù)集和訓(xùn)練集
X_train, X_test = train_test_split(arr,test_size=0.2,random_state=1)

count=10 #k的種類：1、2、3...10
SSE_train=[] #訓(xùn)練集的SSE
SSE_test=[] #測(cè)試集的SSE
n=20 #迭代次數(shù)
for k in range(1,count):
 cla_arr=[] #聚類容器
 centroid=[] #質(zhì)點(diǎn)
 for i in range(k):
 j=np.random.randint(0,len(X_train))
 centroid.append(list1[j])
 cla_arr.append([])
 centroids=np.array(centroid) 
 cla_tmp=cla_arr #臨時(shí)訓(xùn)練集聚類容器
 cla_tmp1=cla_arr #臨時(shí)測(cè)試集聚類容器
 for i in range(n): #開始迭代
 for e in X_train: #對(duì)于訓(xùn)練集中的點(diǎn)進(jìn)行聚類分析
 pi=0
 min_d=distance(e,centroids[pi]) 
 for j in range(k):
 if(distance(e,centroids[j])<min_d): 
  min_d=distance(e,centroids[j])
  pi=j
 cla_tmp[pi].append(e) #添加點(diǎn)到相應(yīng)的聚類容器中
 
 for m in range(k):
 if(n-1==i):
 break
 centroids[m]=means(cla_tmp[m])#迭代更新聚類中心
 cla_tmp[m]=[]
 dis=0
 for i in range(k):  #計(jì)算訓(xùn)練集的SSE_train
 for j in range(len(cla_tmp[i])):
 dis+=distance(centroids[i],cla_tmp[i][j])
 SSE_train.append(dis)
 
 col = ['HotPink','Aqua','Chartreuse','yellow','red','blue','green','grey','orange'] #畫出對(duì)應(yīng)K的散點(diǎn)圖
 for i in range(k):
 plt.scatter([e[0] for e in cla_tmp[i]],[e[1] for e in cla_tmp[i]],color=col[i])
 plt.scatter(centroids[i][0],centroids[i][1],linewidth=3,s=300,marker='+',color='black')
 plt.show()
 
 for e in X_test:  #測(cè)試集根據(jù)訓(xùn)練集的質(zhì)點(diǎn)進(jìn)行聚類分析
 ki=0
 min_d=distance(e,centroids[ki])
 for j in range(k):
 if(distance(e,centroids[j])<min_d):
 min_d=distance(e,centroids[j])
 ki=j
 cla_tmp1[ki].append(e)
 for i in range(k):  #計(jì)算測(cè)試集的SSE_test
 for j in range(len(cla_tmp1[i])):
 dis+=distance(centroids[i],cla_tmp1[i][j])
 SSE_test.append(dis)

SSE=[] #計(jì)算測(cè)試集與訓(xùn)練集SSE的差值
for i in range(len(SSE_test)):
 SSE.append(SSE_test[i]-SSE_train[i])

x=[1,2,3,4,5,6,7,8,9]
plt.figure()
plt.plot(x,SSE_train,marker='*')
plt.xlabel("K")
plt.ylabel("SSE_train")
plt.show() #畫出SSE_train的圖

plt.figure()
plt.plot(x,SSE_test,marker='*')
plt.xlabel("K")
plt.ylabel("SSE_test")
plt.show() #畫出SSE_test的圖

plt.figure()
plt.plot(x,SSE,marker='+')
plt.xlabel("K")
plt.ylabel("SSE_test-SSE_train")
plt.show() #畫出SSE_test-SSE_train的圖