K_average聚类算法的Python简单实现

# -*- coding: utf-8 -*-
#k均值算法 主要部分 简单实现
from random import random


'''
因为单次只能收敛到局部最优解 所以需要多次尝试 最所有对象的误差的平方和最小的结果。
二维数据


'''


#设定K值
K = 3


#从文本中获取对象列表
src_file_name = 'F:\\study\\cluster_algorithm\\k_average\\data.csv'


src_file = open(src_file_name)


data_dict = {}
num = 1
for line in src_file :
    line = line.strip()
    str_list = line.split(',')
    data_dict[num] = (int(str_list[0]) , int(str_list[1]))
    num += 1
    
    
#计算欧氏距离的函数
def get_dist(tp1 , tp2) :
    return ((tp1[0] - tp2[0])**2 + (tp1[1] - tp2[1])**2)**0.5    
    
    


#设置最大尝试次数
max_try = 5


#保存每次的划分结果
cluster_result_dict = {}
#保存每次划分的中心点
cluster_center_dict = {}


#记录每次的误差之和
error_dict = {}


try_cnt = 0
while try_cnt < max_try :
    try_cnt += 1
    #print '1'
    
    center_dict = {}
    #选取最初始中心点
    for i in range(0 , K) :
        x = int(random()*1000000000)%100
        y = int(random()*1000000000)%100
        center_dict[i] = (x , y)
    
    #直到中心点不再改变
    while True :
        #print '2'
        new_center = {}
        #将所有对象指定到欧氏距离最近的中心点上
        cluster_dict = {}
        for obj_id in data_dict :
            #与所有中心点求距离，得到最近的中心点，将其加入到此中心点所代表的簇中
            min_dist = 10000000
            which_cluster = -1 
            for i in range(0 , K) :
                tmp = get_dist(data_dict[obj_id] , center_dict[i])
                if tmp < min_dist :
                    min_dist = tmp
                    which_cluster = i
            if which_cluster == -1 :
                print 'error'
            else :
                if which_cluster not in cluster_dict :
                    cluster_dict[which_cluster] = []
                cluster_dict[which_cluster].append(obj_id)
        #计算新的中心点
        for i in range(0 , K) :
            sum_x = 0
            sum_y = 0
            cnt = 0
            for obj_id in cluster_dict[i] :
                sum_x += data_dict[obj_id][0]
                sum_y += data_dict[obj_id][1]
                cnt += 1
            avg_x = (sum_x + 0.0)/cnt
            avg_y = (sum_y + 0.0)/cnt
            
            new_center[i] = (avg_x , avg_y)
        
        #比较新旧中心点
        break_flag = 1
        for i in range(0 , K) :
            if new_center[i] != center_dict[i] :
                break_flag = 0    
            
        #如果中心点没有再改变
        if break_flag == 1 :
            #保存此时的簇
            cluster_result_dict[try_cnt] = cluster_dict
            cluster_center_dict[try_cnt] = new_center
            break
        #如果发生了改变，更新中心中
        else :
            for i in range(0 , K) :
                center_dict[i] = new_center[i]
            
    #计算簇内变差
    sum_error = 0
    for i in range(0 , K) :
        print i
        for obj_id in cluster_result_dict[try_cnt][i] :
            dist = get_dist(data_dict[obj_id] , cluster_center_dict[try_cnt][i])
            sum_error += dist**2
    error_dict[try_cnt] = sum_error
    
    
base_dir = 'F:\\study\\cluster_algorithm\\k_average\\result\\'
for i in cluster_result_dict :
    file_name = base_dir + str(i) + '.txt'
    file_tmp = open(file_name , 'w')
    
    str_buffer = ''
    for j in cluster_result_dict[i] :
        str_buffer += 'cluster ' + str(j) + '\n'
        for obj_id in cluster_result_dict[i][j] :
            str_buffer += str(data_dict[obj_id][0]) + ',' + str(data_dict[obj_id][1]) + '\n'
    
    file_tmp.write(str_buffer)
    file_tmp.close()