简单的Kmeans实现

import numpy as np
import random
import pandas as pd
import matplotlib.pyplot as plt

#K = 4 需要的质心数量
# 初始化数据
def init_data():
    df = pd.read_csv('/mnt/d/EDA/Cluster/cluster_data.csv')
    nd = np.array(df)
    # 归一化
    data = (nd - np.min(nd,axis=0)) / (np.max(nd,axis=0) - np.min(nd,axis=0))
    return data

# 距离使用欧几里得距离
def euclidean_metric(vector,centre):
    distance = np.sqrt(np.sum(np.square(vector-centre),axis=1))
    return distance

# 确定初始质心
def init_centre(data, K):
    z = np.zeros((K,data.shape[1]))
    for i in range(K):
        x = random.randint(0,len(data)) #随机产生一个索引值，通过索引得到一个质心的位置矩阵
        z[i] = data[x]
    return z

# 开始计算距离质心最小距离
def K_means(K=4):
    data = init_data()
    z = init_centre(data, K)
    centre_all = [np.zeros((K,data.shape[1]))]
    centre_all.append(z)
    run_time = 0
    # 循环新的质心并计算距离，将数据重新划分到簇中
    while ((centre_all[-1] == centre_all[-2]).all()==False) and (run_time<100):
    # for y in range(10):
        li = []
        for i in data:
            li.append(euclidean_metric(i,z))
        li = np.array(li)

        group = []
        for i in range(0,len(li)):
            point = [0,0]
            point[0] = i     # index in data
            point[1] = np.argmin(li[i]) # belong to which cluster(group)
            group.append(point)
        group = np.array(group)

        # 根据划分出来的簇的中心更新质心
        new_c = []
        for i in range(K):
            t = data[group[:,1]==i]    # numpy可以使用布尔值来进行索引
            new_centre = np.mean(t,axis=0)
            new_c.append(new_centre)
        z = np.array(new_c)
        centre_all.append(z)
        run_time += 1

    # 绘制图像
    colorlist = ['b','r','g','purple'] #对每一个簇都用不同的颜色标记
    plt.figure()
    for i in range(4):
        t = data[group[:,1]==i]
        plt.scatter(t[:,0],t[:,1],c=colorlist[i])
    plt.show()