机器学习必刷题-手撕推导篇(1)

def kmeans(k):
    m, n = 100, 20  # 构造样本：100行、20列
    x = 10 * np.random.random((m, n))

    # 随机选择k个初始中心点
    init_cent_sample = set()
    while len(init_cent_sample) < k:
        init_cent_sample.add(np.random.randint(0, m))
    cent = x[list(init_cent_sample)]

    # 记录每个样本的类归属
    cluster_assessment = np.zeros((m, 2))

    # 记录每个类的中心点在本次迭代后是否有过改变
    cent_changed = True
    while cent_changed:
        cent_changed = False

        for j in range(m):
            # 记录每个样本距离最近的类
            min_inx = -1
            # 记录每个样本的最小类距
            min_dist = math.inf

            for i in range(k):
                d = distance(x[j], cent[i])
                if d < min_dist:
                    min_dist = d
                    min_inx = i

            # 记录此样本的中心点是否发生变化
            if min_inx != cluster_assessment[j][0]:
                cluster_assessment[j] = np.array([min_inx, min_dist])
                cent_changed = True
        print(cluster_assessment)

        # 更新每个类的中心点：均值
        for i in range(k):
            cent_i_samples = np.where(cluster_assessment[:, 0] == i)
            if len(cent_i_samples) > 0:
                print(cent_i_samples)
                cent[i] = np.mean(x[cent_i_samples], axis=0)


# 计算距离
def distance(a, b):
    return math.sqrt(sum(pow(a - b, 2)))