Python手动实现k-means

import numpy as np

import matplotlib.pyplot as plt

def kmeans(data, cluster_num, method='mean'):

    points = np.array(data)

    labels = []

    random_idx = []

    while True:

        t = np.random.randint(points.shape[0])

        if t not in random_idx:

            random_idx.append(t)

        if len(random_idx) == cluster_num:

            break

    seeds = []

    for i in range(cluster_num):

        seeds.append(points[random_idx[i]])

    seeds = np.array(seeds)

    prev_seeds = seeds + 10

    colors = ["red", "blue", "yellow", "cyan", "purple"]

    points_labeled = []

    for i in range(cluster_num):

        points_labeled.append([])

    while sum(abs((prev_seeds - seeds).ravel())) > 0:

        prev_seeds = seeds.copy()

        for i in range(points.shape[0]):

            t_min = 10000

            for j in range(seeds.shape[0]):

                t_current = np.sqrt((points[i][0] - seeds[j][0]) ** 2 + (points[i][1] - seeds[j][1]) ** 2)

                if t_current < t_min:

                    t_min = t_current

                    try:

                        labels.pop(i)

                    except IndexError:

                        pass

                    labels.append(j)

        for i in range(len(labels)):

            points_labeled[labels[i]].append(points[i].tolist())

        points_labeled = np.array(points_labeled)

        for i in range(points_labeled.shape[0]):

            if points_labeled[i]:

                plt.scatter(np.array(points_labeled[i])[:, 1], np.array(points_labeled[i])[:, 0], c=colors[i])

        for i in seeds:

            plt.scatter(i[1], i[0], c="black", linewidths=3)

        plt.show()

        for i in range(seeds.shape[0]):

            if method == 'mean':

                seeds[i] = np.array(

                    [np.mean(np.array(points_labeled[i])[:, 0]), np.mean(np.array(points_labeled[i])[:, 1])])

            elif method == 'median':

                seeds[i] = np.array(

                    [np.median(np.array(points_labeled[i])[:, 0]), np.median(np.array(points_labeled[i])[:, 1])])

        points_labeled = []

        for i in range(cluster_num):

            points_labeled.append([])

        labels = []

    return seeds, labels, sum(abs((prev_seeds - seeds).ravel()))

if __name__ == "__main__":

    points = [

        [2, 4], [4, 2], [6, 2], [5, 3], [5, 5], [7, 5],

        [5, 15], [6, 17], [4, 14], [5, 13], [9, 15], [3, 14], [7, 13],

        [20, 16], [19, 15], [17, 15], [16, 14], [14, 18], [22, 10], [17, 17], [16, 13], [18, 14], [17, 13],

        [22, 26], [24, 23], [25, 25], [26, 22], [26, 26], [26, 28], [28, 18], [28, 28]

    ]

    cluster_num = 4

    [centroids, labels, interia] = kmeans(points, cluster_num)
巴特西

Python手动实现k-means

最新文章

热门文章
