机器学习系列之聚类

三种常用的聚类算法：

    (1)基本K均值：基于原型的，划分的聚类技术，试图从全部数据对象中发现用户指定个数的簇。

    (2)凝聚层次聚类：开始每个点各成一簇，然后重复的合并两个最近的簇，直到指定的簇个数。

    (3)DBSCAN：一种划分的，基于密度的聚类算法。

优缺点，不再赘述，参考https://www.cnblogs.com/giserliu/archive/2015/04/05/4394807.html

#!/usr/bin/env python
# -*- coding:utf-8 -*-
__author__ = 'Great'
"""
指定三个质心
计算每个点到三个质心的距离
指派每个点到簇
更新质心
记录质心
显示分类及质心轨迹
"""
import numpy as np
a = np.random.randint(1,100, 80, dtype = 'int64')
b = np.random.randint(1, 150, 80)
points = []
for i in range(80):
    points.append((a[i], b[i]))

import pylab as pl

#指定质心
current_point1 = [22, 111]
current_point2 = [100, 12]
current_point3 = [50, 56]
#显示质心
pl.plot([current_point1[0]], [current_point1[1]], 'ok')
pl.plot(current_point2[0], current_point2[1], 'ok')
pl.plot(current_point3[0], current_point3[1], 'ok')

#记录质心轨迹
current1 = [current_point1]
current2 = [current_point2]
current3 = [current_point3]

#三个聚类簇
group1 = []
group2 = []
group3 = []

for cost_time in range(100):
    group1 = []; group2 = []; group3 = []

    for onepoint in points:
        distance1 = pow(abs(onepoint[0]-current_point1[0]), 2) + pow(abs(onepoint[1]-current_point1[1]), 2)
        distance2 = pow(abs(onepoint[0] - current_point2[0]), 2) + pow(abs(onepoint[1] - current_point2[1]), 2)
        distance3 = pow(abs(onepoint[0] - current_point3[0]), 2) + pow(abs(onepoint[1] - current_point3[1]), 2)

        #指派到最近的簇
        min_len = min(distance1, distance2, distance3)
        if min_len == distance1:
            group1.append(onepoint)
        if min_len == distance2:
            group2.append(onepoint)
        if min_len == distance3:
            group3.append(onepoint)

    #更新质心
    current_point1 = [sum([onepoint[0] for onepoint in group1])/len(group1), sum([onepoint[1] for onepoint in group1])/len(group1)]
    current_point2 = [sum([onepoint[0] for onepoint in group2])/len(group2), sum([onepoint[1] for onepoint in group2])/len(group2)]
    current_point3 = [sum([onepoint[0] for onepoint in group3])/len(group3), sum([onepoint[1] for onepoint in group3])/len(group3)]

    current1.append(current_point1)
    current2.append(current_point2)
    current3.append(current_point3)
#打印簇
pl.plot([onepoint[0] for onepoint in group1], [onepoint[1] for onepoint in group1], "or")
pl.plot([onepoint[0] for onepoint in group2], [onepoint[1] for onepoint in group2], "oy")
pl.plot([onepoint[0] for onepoint in group3], [onepoint[1] for onepoint in group3], "og")

#打印质心轨迹
for center in [current1, current2, current3]:
    pl.plot([eachcenter[0] for eachcenter in center], [eachenter[1] for eachenter in center])
#显示
pl.show()
#输出
print(current_point1, current_point2, current_point3)
print(group1)
print(group2)
print(group3)

本文，只是实现了基本的K均值聚类算法，并且没有对聚类结果进行优化。只是熟悉下，相关的计算过程。其中样本集，为随机生成的。过程主要参照https://www.oschina.net/code/snippet_176897_14731

希望自己越来越好，加油！走在成长的路上。