聚类--K均值算法:自主实现与sklearn.cluster.KMeans调用

import numpy as np
x = np.random.randint(1,50,[20,1])
y = np.zeros(20)
k = 3
#1) 选取数据空间中的K个对象作为初始中心,每个对象代表一个聚类中心;
def initcen(x,k):
    return x[:k]
#2) 对于样本中的数据对象,根据它们与这些聚类中心的欧氏距离,按距离最近的准则将它们分到距离它们最近的聚类中心(最相似)所对应的类;
def nearest(kc,i):
    d = abs(kc-i)
    w = np.where(d == np.min(d))
    return w[0][0]
 
def xclassify(x,y,kc):
    for i in range(x.shape[0]):
        y[i] = nearest(kc,x[i])
        return y
 
#3) 更新聚类中心:将每个类别中所有对象所对应的均值作为该类别的聚类中心,计算目标函数的值;
 
def kcmean(x,y,kc,k):
    l = list(kc)
    flag = False
    for c in range(k):
        m = np.where(y ==0)
        n = np.mean(x[m])
        if l[c] != n:
            l[c] = n
            flag = True
            print(l,flag)
    return (np.array(l),flag)
#4) 判断聚类中心和目标函数的值是否发生改变,若不变,则输出结果,若改变,则返回2)
kc = initcen(x,k)
 
flag = True
print(x,y,kc,flag)
while flag:
    y = xclassify(x,y,kc)
    kc,flag = kcmean(x,y,kc,k)
print(y,kc)

from sklearn.datasets import load_iris
iris = load_iris()
datas = iris.data
iris_length=datas[:,2]
 
# 用鸢尾花花瓣作分析
x = np.array(iris_length)
y = np.zeros(x.shape[0])
kc = initcen(x,3)
flag = True
while flag:
    y = xclassify(x,y,kc)
    kc,flag = kcmean(x,y,kc,3)
print(kc,flag)
 
# 分析鸢尾花花瓣长度的数据,并用散点图表示出来
import matplotlib.pyplot as plt
plt.scatter(iris_length, iris_length, marker='p', c=y, alpha=0.5, linewidths=4, cmap='Paired')
plt.show()

from sklearn.cluster import KMeans
 
iris_length = datas[:, 2:3]
k_means = KMeans(n_clusters=3)
result = k_means.fit(iris_length)
kc1 = result.cluster_centers_
y_kmeans = k_means.predict(iris_length)
 
# 画图
plt.scatter(iris_length,np.linspace(1,150,150),c=y_kmeans,marker='x',cmap='rainbow',linewidths=4)
plt.show()

k_means1 = KMeans(n_clusters=3)
result1 = k_means1.fit(datas)
kc2 = result1.cluster_centers_
y_kmeans1 = k_means1.predict(datas)
 
print(y_kmeans1, kc2)
print(kc2.shape, y_kmeans1.shape, datas.shape)
 
plt.scatter(datas[:, 0], datas[:, 1], c=y_kmeans1, marker='p', cmap='flag', linewidths=4, alpha=0.6)
plt.show()

posted on 2018-12-24 10:43  zoyeln  阅读(225)  评论(0编辑  收藏  举报

导航