1.用python实现K均值算法
K-means是一个反复迭代的过程,算法分为四个步骤:
(x,k,y)
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 initcenter(x, k): kc
import numpy as np x = np.random.randint(1,100,[20,1]) y = np.zeros(20) k = 3 def initcenter(x, k): return x[:k] kc = initcenter(x,k) kc
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, x[i]): j
def xclassify(x, y, kc):y[i]=j
d = abs(kc - 78) np.where( d == np.min(d))[0][0] def initcenter(x, k): return x[:k] def nearest(kc, i): d = (abs(kc - i)) w = np.where(d == np.min(d)) return w[0][0] kc = initcenter(x,k) nearest(kc,93) def initcenter(x, k): return x[:k] 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 kc = initcenter(x,k) y = xclassify(x, y, kc) print(kc,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):
import numpy as np from sklearn.datasets import load_iris iris = load_iris() x = iris.data[:,1] y = np.zeros(150) def initcenter(x, k): return x[:k].reashape(k) def nearest(kc, i): d = (abs(kc - i)) w = np.where(d == np.min(d)) return w[0][0] def kcmean(x, y, kc, k): l = list(kc) flag = False for c in range(k): m = np.mean(x[m]) if m[0].shape !=(0,): n = np.mean(x[m]) if l[c] !=n: l[c] = n flag = True return (np.array(l),flag) def xclassify(x, y, kc): for i in range(x.shape[0]): y[i] = nearest(kc, x[i]) return y k = 3 kc = initcent(x, k) flag = True print(x, y, kc, k) kc = initcenter(x,k): y = xclassify(x, y, kc) kc, flag = kcmean(x, y, kc, k)
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)。
while flag:
y = xclassify(x, y, kc)
kc, flag = kcmean(x, y, kc, k)
while flag: y = xclassify(x,y,kc) kc, flag = kcmean(x,y,kc,k) print(y,kc,type(kc)) print(x,y)
2. 鸢尾花花瓣长度数据做聚类并用散点图显示。
import matplotlib.pyplot as plt plt.scatter(x,x,c=x,s=50,cmap="rainbow"); plt.show()
3. 用sklearn.cluster.KMeans,鸢尾花花瓣长度数据做聚类并用散点图显示.
import numpy as np from sklearn.datasets import load_iris iris=load_iris() x=iris.data[:,1] y=np.zeros(150) def initcenter(x,k): #初始聚类中心数组 return x[0:k].reshape(k) def nearest(kc,i): #数组中的值,与聚类中心最小距离所在类别的索引号 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]): #对数组的每个值进行分类,shape[0]读取矩阵第一维度的长度 y[i] = nearest(kc,x[i]) return y def kcmean(x,y,kc,k): #计算各聚类新均值 l = list(kc) flag = False for c in range(k): print(c) m = np.where(y == c) n=np.mean(x[m]) if l[c] != n: l[c] = n flag = True #聚类中心发生变化 print(l,flag) return (np.array(l),flag) k = 3 kc = initcenter(x,k) flag = True print(x,y,kc,flag)
4. 鸢尾花完整数据做聚类并用散点图显示
while flag: y = xclassify(x,y,kc) kc, flag = kcmean(x,y,kc,k) print(y,kc,type(kc)) print(x,y) import matplotlib.pyplot as plt plt.scatter(x,x,c=y,s=50,cmap="Paired"); plt.show() #用sklearn.cluster.KMeans,鸢尾花花瓣长度数据做聚类并用散点图显示 import matplotlib.pyplot as plt import numpy as np from sklearn.datasets import load_iris iris = load_iris() X = iris.data X from sklearn.cluster import KMeans est = KMeans(n_clusters = 3) est.fit(X) kc = est.cluster_centers_ y_kmeans = est.predict(X) #预测每个样本的聚类索引 print(y_kmeans,kc) print(kc.shape,y_kmeans.shape,np.shape) plt.scatter(X[:,0],X[:,1],c=y_kmeans,s=50,cmap='rainbow'); plt.show()