# -*-coding: utf-8 -*-

# 算法实现流程：

# 创建k个点作为起始质心（经常是随机选择）

# 当任意一个点的簇分配发生改变时

# 对数据集中的每个数据点

# 对每个质心

# 计算质心与数据点之间的距离

# 将数据点分配到距其最近的簇

# 对每一个簇，计算簇中所有点的均值并将均值作为质心

import kMeans

from numpy import*

import matplotlib

import matplotlib.pyplot as plt

import time

st = time.time()

k = 3

dataMat = mat(kMeans.loadDataSet('testSet2.txt'))

oldClassLabel = zeros(len(dataMat),int)

newClassLabel = ones(len(dataMat),int)

center = kMeans.randCent(dataMat, k)

dist = []

m = 0

while newClassLabel.tolist().__eq__(oldClassLabel.tolist()) != True: # 所有的点的新分类的标签和旧分类的标签不一致时就继续进行划分

m += 1 # 迭代次数

for di in range(len(dataMat)): # 对数据集中的每个数据点

dist = []

for ci in range(len(center)): # 对每个质心，计算某个点到质心的距离

dist.append(kMeans.distEclud(dataMat[di], center[ci])) # dist 用于记录一个点到所有簇点的距离

distsort = array(dist).argsort() # 对距离排序，返回从小到大的索引

oldClassLabel = newClassLabel.copy() # ！！！！注意这里是引用不能直接用等号，否则将会使得两个值一起变

newClassLabel[di] = distsort[0] # 取出索引的最小值，就是距离最近的点

for j in range(k):

x = mean(array(dataMat)[kMeans.find_all_index(newClassLabel, j), 0]) # 重新计算簇心的坐标

y = mean(array(dataMat)[kMeans.find_all_index(newClassLabel, j), 1])

center[j] = [x, y]

fig = plt.figure()

ax = fig.add_subplot(111)

center = array(center)

ax.scatter(center[:,0],center[:,1],marker = '+' ,c='r')

print center

print m

dataMat = array(dataMat)

et = time.time()

print 'The total time is:',et-st

ax.scatter(dataMat[kMeans.find_all_index(newClassLabel, 0),0],dataMat[kMeans.find_all_index(newClassLabel, 0),1],c = 'b')

ax.scatter(dataMat[kMeans.find_all_index(newClassLabel, 1),0],dataMat[kMeans.find_all_index(newClassLabel, 1),1],c = 'g')

ax.scatter(dataMat[kMeans.find_all_index(newClassLabel, 2),0],dataMat[kMeans.find_all_index(newClassLabel, 2),1],c = 'c')

ax.scatter(dataMat[kMeans.find_all_index(newClassLabel, 3),0],dataMat[kMeans.find_all_index(newClassLabel, 3),1],c = 'k')

plt.show()