K-means和谱聚类 ——python实现

文件目录

data5用于K-means
data2用于谱聚类

K-means 实现

算法步骤:

step1:初始化聚类中心
step2:计算每个样本点与聚类中心的距离，将其归为最近的一类
step3:计算聚类之后的每个类别的均值作为新的类别中心
step4:迭代进行 step2，step3，直到类别中心不再发生变化为止
step5:计算准确率

代码

import numpy as np
import matplotlib.pyplot as plt

def draw(data,c):
    color = ['r', 'b', 'k', 'g', 'm']
    for i in range(c):
        plt.scatter(data[data[:, 2] == i + 1, 0],\
                    data[data[:, 2] == i + 1, 1],c=color[i])
    plt.show()


def distance(x1,x2):
    return np.sqrt(np.sum(np.power(x2-x1,2)))


def kmeans(data,c):
    row,clm=data.shape
    u = []
    d = np.zeros(data.shape[0])
    d[:] = 1000
    Y = data.copy()
    for i in range(c):
        #u.append(data[np.random.randint(0, high=row, size=1),0:2])
        #u.append(data[i,0:2])
        u.append(data[i*(row//c),0:clm-1])
    u = np.array(u)
    u_ = np.zeros(u.shape)
    while np.sum(np.abs(u-u_)) != 0:
        for i in range(data.shape[0]):
            for j in range(u.shape[0]):
                if distance(data[i, 0:clm-1], u[j]) < d[i]:
                    Y[i, -1] = j+1
                    d[i] = distance(data[i, 0:clm-1], u[j])
        u_ = u.copy()
        for j in range(c):
            u[j] = np.mean(Y[Y[:, -1] == j+1,0:clm-1], axis=0)
    return u,Y


def acc(data,Y,c):
    num_c = int(data.shape[0] / c)
    num = np.zeros([c,c])
    for i in range(c):
        for j in range(c):
            num[i,j] = np.sum(Y[num_c*i:num_c*i+num_c, -1] == j+1)

    count = np.sum(np.max(num,axis=1))
    return count/data.shape[0]


def dis(data,u,c):
    u_ = np.zeros(u.shape)
    num = np.zeros([c, c])
    for j in range(c):
        u_[j] = np.mean(data[data[:, -1] == j+1,0:2], axis=0)
    for i in range(c):
        for j in range(c):
            num[i,j] = distance(u[i],u_[j])

    count = np.min(num,axis=1)
    return count


if __name__ == '__main__':
    data = np.loadtxt("data5.txt")
    draw(data, 5)
    u, Y = kmeans(data, 5)
    draw(Y, 5)
    ac = acc(data, Y, 5)
    print("类别中心：", u)
    print('均值距离：', dis(data,u,5))
    print("精度：", ac)

测试

谱聚类

原理

代码

import numpy as np
import matplotlib.pyplot as plt
import Kmeans

def draw1(data,Y,c):
    color = ['r', 'b', 'k', 'g', 'm']
    for i in range(c):
        plt.scatter(data[Y[:, -1] == i + 1, 0],\
                    data[Y[:, -1] == i + 1, 1],c=color[i])
    plt.show()

def graph(data,k,sigma,c):
    g = np.zeros([data.shape[0],data.shape[0]])
    for i in range(data.shape[0]):
        for j in range(data.shape[0]):
            g[i,j] = np.exp(-np.sum((data[i]-data[j]) ** 2)/(2*(sigma**2)))
            if i==j:
                g[i,j]=0

    g_k = np.zeros(g.shape)
    for i in range(k):
        for j in range(data.shape[0]):
            g_k[j,np.argmax(g[j])] = np.max(g[j])
            g_k[np.argmax(g[j]),j] = g_k[j,np.argmax(g[j])]
            g[j,np.argmax(g[j])] = 0
    g_k = (g_k.T+g_k)/2
    d_k = np.sum(g_k,axis=1)
    L = np.diag(d_k) - g_k

    sq = np.diag(1.0/(d_k ** 0.5))
    L = np.dot(np.dot(sq,L),sq)

    A, B = np.linalg.eig(L)
    feature = B[:, np.argsort(A)[:c]].copy()
    data_f =np.c_[feature,np.zeros(feature.shape[0])]
    u, Y = Kmeans.kmeans(data_f,2)
    return u, Y

if __name__ == '__main__':
    data = np.loadtxt("data2.txt")
    # data = data[1:5,:]
    c = np.zeros(data.shape[0])
    ac = []
    c[0:100] = 1
    c[100:200] = 2
    data = np.c_[data,c]
    Kmeans.draw(data,2)
    for k in range(1,10):
        u, Y = graph(data,k,2,2)
        ac.append(Kmeans.acc(data, Y, 2))
    plt.xlabel("K")  # x轴上的名字
    plt.ylabel("ACC")  # y轴上的名字
    plt.title('sigma=2')
    plt.plot(range(1,10),ac)
    plt.show()
    ac1 = []
    for i,sigma in enumerate(np.arange(0.5,4,0.5)):
        print(sigma)
        u, Y = graph(data,4,sigma,2)
        ac1.append(Kmeans.acc(data, Y, 2))
    plt.xlabel("sigma")  # x轴上的名字
    plt.ylabel("ACC")  # y轴上的名字
    plt.title('K=4')
    plt.plot(np.arange(0.5,4,0.5),ac1)
    plt.show()

测试