点云学习【3.1】k-means聚类算法

K-means聚类步骤：
1.随机选取K个中心点；
2.每个数据点分配给K个中心点；
3.通过每个类的均值重新计算中心点；
4.对步骤2-3进行迭代计算。

import numpy as np
import random

class K_means():
    def __init__(self,n_clusters=2,tolerance=0.0001,max_iter=300):
        #n_clusters是k（聚类数量），tolerance是允许误差，max_iter是迭代次数
        self.k_ = n_clusters
        self.tolerance = tolerance
        self.max_iter_=max_iter
    def fit(self,data):
        #fit分为Estep和Mstep，分别计算上述步骤2-3.
        #random.sample(a,b)是指把返回b个a范围内的索引
        centers = data[random.sample(range.shape[0],self.k_)]
        old_centers = np.copy(centers)
        labels = [[] for i in range(self.k_)]
        for iter_ in range(self.max_iter_):
            for idx,point in enumerate(data):
                diff = np.linalg.norm(old_centers-point,axis=1)
                labels[argmin(diff)].append(idx)
            for i in range(self.k_):
                points = data[labels[i],:]
                centers[i] = np.mean(points,axis=0)
            if np.sum(np.abs(old_centers-centers))<self.tolerance*self.k_:
                break
            old_centers = np.copy(centers)
        self.centers = centers
        self.fitted = True


    def predict(self,p_data):
        result = []
        if not self.fitted:
            print("unfitted")
            return result
        for point in p_data:
            diff = np.linalg.norm(self.centers-point,axis=1)
            result.append(argmin(diff))
        return result