k_means简单实现-K值选择

k_means

1、一维数据聚类：

def k_means(cen_num:int, x):
    cen = torch.rand(cen_num, 1)
#     x = torch.tensor(x, dtype=torch.float32)
#     cen = torch.tensor(cen, dtype=torch.float32)
    
    for i in range(50):
        #SSE, 误差平方和
        sse = (x - cen)**2
        #取最后一行作为标签
        values, label = torch.sort(sse, 0)
        for i in range(cen_num):
            index = torch.where(label[0]==i)[0]
            cen[i] = torch.mean(x[index])
    return torch.sum(values[0]).item()
sse = []
x = torch.rand(1, 10000)
for i in range(2, 50):
    sse.append(k_means(i, x))
    plt.plot(sse)
    plt.show()
    plt.pause(0.1)

2、二维数据（一维数据类似）：

def k_means2dim(cen_num:int, x, iterations):
    cen = torch.rand(cen_num,1, 2)
    for j in range(iterations):
        sse = (x - cen) **2
        sse = sse.sum(axis=(2))	#输入数据与中心点距离的平方
        values, label = torch.sort(sse, dim=0) #按零轴排序=分类
        for i in range(cen_num):
            index = torch.where(label[0]==i)[0]
            cen[i] = torch.mean(x[index], axis=0) #获取新中心点
    return torch.sum(values[0]).item()
   
sse = []
x = torch.rand(10000, 2)
for i in range(2, 50):
    sse.append(k_means2dim(i, x, 50))
    plt.plot(sse)
    plt.show()
    plt.pause(0.1)