python实现K-means聚类算法

python实现Kmeans

K-Means算法是典型的基于距离的聚类算法，采用距离作为相似性的评价指标，两个样本点的距离越近，其相似度就越大。在K-means算法中，cluster是由距离靠近的样本点组成的，因此把得到距离接近的cluster作为目标。
步骤：

• 指定k个初始质心(centroids)，以作为聚类的初始cluster
• 对所有点计算到各个质心的距离，将该点类别标为最近质心的cluster
• 重新计算k个cluster的新质心
• 重复计算，直到算法收敛（质心不再发生变化）

import pandas as pd
import numpy as np
import data_utils as du
from scipy.spatial.distance import pdist

# 计算距离
def dist(x, y, p=1):
    X = np.vstack([x, y])
    dist = float("inf")
    # L1范式
    if p == 1:
        dist = np.linalg.norm(x - y, 1)
    # 欧式距离  L2范式
    if p == 2:
        # dis = np.sqrt(np.sum(np.square(x-y)))
        dist = np.linalg.norm(x - y, 2)
    # 标准化欧氏距离
    elif p == 3:
        dist = pdist(X, 'seuclidean')
    # cosin 余弦相似度
    elif p == 4:
        dist = pdist(X,'cosine')
    # 马氏距离
    elif p == 5:
        dist = pdist(X, 'mahalanobis')
    return dist

    
# 构建聚簇中心，取k个随机距离质心
def randCent(data, k):
    cents = np.array(random.sample(list(data), k))
    return cents

# pdist：1.L1范式 2.欧式距离 3.标准化欧式距离 4.余弦相似度 5.马氏距离
def kMeans(data, k, pdist):
    rn = data.shape[0]
    cluster = np.zeros(rn)
    cents = randCent(data, k) # 选取k个初始质心（作为初始cluster）
    iterations = 0 # 迭代次数
    
    while True:	# 迭代
        iterations += 1
        # print('iterations=', iterations)
        if iterations > 1000: break
        for i in range(rn):	# 把每一个数据点划分到离它最近的中心点
            m_dist = np.inf
            m_index = -1	# 属于第几个聚簇
            for j in range(k):
                d = dist(cents[j, :], data[i, :], pdist)	# i点到j中心的距离
                if d < m_dist:
                    m_dist = d
                    m_index = j
            cluster[i] = m_index
        new_cents = cents.copy()
        for cent_i in range(k):  # 重新计算中心点
            k_cluster = data[np.nonzero(cluster == cent_i)] # 找到第cent_i聚簇的所有点
            if len(k_cluster) != 0:
                new_cents[cent_i, :] = np.mean(k_cluster, 0)  # 算出这些数据的中心点
        # print('cents=', cents)
        # print('new_cents=', new_cents)
        # print(new_cents == cents)
        if (new_cents == cents).all():	# 迭代结束
            break
        cents = new_cents.copy()
    return cluster, iterations

if __name__ =='__main__':
    print("----k_means----")
    k = 4
    pdist = 1
    data, labels = du.read_data()
    label_set = sorted(list(set(labels)))
    cluster, iterations = k_means(data, k, pdist)
    print("end of k_means \niterations=", iterations)
    # print("----------------")

data_utils.py

import pandas as pd
import numpy as np

def read_data():
    pf = pd.read_csv('Frogs_MFCCs.csv', encoding='utf-8')
    data = np.array(pf)
    f_data = data[:, 0:22].astype(np.float)
    return f_data, data[:, 22]

if __name__ == '__main__':
    f_data, labels = read_data()
    label_set = list(set(labels))

代码及数据下载：https://download.csdn.net/download/SAM2un/12036441