K-means 简单实现-python

%matplotlib inline

from sklearn.datasets import make_blobs
from matplotlib import pyplot
import numpy as np
import pandas as pd
import random

#使用make_blobs 产生测试数据
data,target = make_blobs(n_samples=100,n_features=2,centers=3)
pyplot.scatter(data[:,0],data[:,1],c=target)
pyplot.show()

#定义各种距离
#欧式距离
def eu_distance(x,y):
    return np.sqrt(sum(np.square(x-y)))

#普通k_means，结束条件为迭代次数
def _k_means_normal(data,iter_num,center_num,dis):
    print(len(data))
    #随机选择质心
    print('初始化质心')
    indexs = list(range(len(data)))
    random.shuffle(indexs)
    init_centroids_index = indexs[:center_num]
    centroids = data[init_centroids_index,:]
    #确定种类编号
    levels = list(range(center_num))  
    print('分配种类编号')
    print('开始迭代')
    sample_target=[]
    for i in range(iter_num):
        
        new_centroids=[[] for i in range(center_num)]
        new_centroids_num=[0 for i in range(center_num)]
        sample_target=[]
        
        #遍历数据
        for sample in data:
            #计算距离，由距离该数据最近的质心，确定该点所属类别
            distances = [dis(sample,centroid) for centroid in centroids]
            cur_level = np.argmin(distances)  
            sample_target.append(cur_level)
            
            #统计，方便迭代完成后重新计算质心
            new_centroids_num[cur_level]+=1
            if len(new_centroids[cur_level]) < 1:
                new_centroids[cur_level] = sample
            else:
                new_centroids[cur_level] = new_centroids[cur_level]+sample
        
        #重新划分质心
        centroids = []
        for centroid, num in zip(new_centroids,new_centroids_num):
            centroids.append([item/num for item in centroid])
        centroids = np.array(centroids)
        
    print('结束')
#     print(sample_target)
    return sample_target 

def k_means(data,iter_num=10,center_num=3,cls='normal',dis=eu_distance):
    if cls == 'normal':
        return _k_means_normal(data,iter_num,center_num,dis=dis)
    

predict = k_means(data,iter_num=200)
pyplot.scatter(data[:,0],data[:,1],c=predict)