# -*- coding=utf-8 -*-

"""
authon:xuwf
created:2017-02-07
purpose:实现k-means算法
"""

import numpy as np
import random

'''装载数据'''
def load():
 data=np.loadtxt('data\k-means.csv',delimiter=',')
 return data

'''计算距离'''
def calcDis(data,clu,k):
 clalist=[] #存放计算距离后的list
 data=data.tolist() #转化为列表
 clu=clu.tolist()
 for i in range(len(data)):
  clalist.append([])
  for j in range(k):
   dist=round(((data[i][1]-clu[j][0])**2+(data[i][2]-clu[j][1])**2)*0.05,1)
   clalist[i].append(dist)
 clalist=np.array(clalist) #转化为数组
 return clalist

'''分组'''
def group(data,clalist,k):
 grouplist=[] #存放分组后的集群
 claList=clalist.tolist()
 data=data.tolist()
 for i in range(k):
  #确定要分组的个数，以空列表的形式，方便下面进行数据的插入
  grouplist.append([])
 for j in range(len(clalist)):
  sortNum=np.argsort(clalist[j])
  grouplist[sortNum[0]].append(data[j][1:])
 grouplist=np.array(grouplist)
 return grouplist

'''计算质心'''
def calcCen(data,grouplist,k):
 clunew=[]
 data=data.tolist()
 grouplist=grouplist.tolist()
 templist=[]
 #templist=np.array(templist)
 for i in range(k):
  #计算每个组的新质心
  sumx=0
  sumy=0
  for j in range(len(grouplist[i])):
   sumx+=grouplist[i][j][0]
   sumy+=grouplist[i][j][1]
  clunew.append([round(sumx/len(grouplist[i]),1),round(sumy/len(grouplist[i]),1)])
 clunew=np.array(clunew)
 #clunew=np.mean(grouplist,axis=1)
 return clunew

'''优化质心'''
def classify(data,clu,k):
 clalist=calcDis(data,clu,k) #计算样本到质心的距离
 grouplist=group(data,clalist,k) #分组
 for i in range(k):
  #替换空值
  if grouplist[i]==[]:
   grouplist[i]=[4838.9,1926.1]
 clunew=calcCen(data,grouplist,k)
 sse=clunew-clu
 #print "the clu is :%r\nthe group is :%r\nthe clunew is :%r\nthe sse is :%r" %(clu,grouplist,clunew,sse)
 return sse,clunew,data,k 

if __name__=='__main__':
 k=3 #给出要分类的个数的k值
 data=load() #装载数据
 clu=random.sample(data[:,1:].tolist(),k) #随机取质心
 clu=np.array(clu)
 sse,clunew,data,k=classify(data,clu,k)
 while np.any(sse!=0):
  sse,clunew,data,k=classify(data,clunew,k)
 clunew=np.sort(clunew,axis=0)
 print "the best cluster is %r" %clunew