K Means Algorithm Implementation

k_means.py

import matplotlib.pyplot as plt
import numpy as np
from scipy.spatial import ConvexHull


class KMeans:
    mean_vec = np.array([])
    X = np.array([])
    y = np.array([])
    clusters = []

    def fit(self, X, k, visual=False, epsilon=0.0001):
        self.X = X
        # Initialize mean vectors
        self.mean_vec = X[np.random.randint(0, len(X), k)]
        stop = False
        iter = 0
        while not stop:
            stop = True
            # Clustering
            self.y = np.empty([len(X)])
            for i in range(0, len(X)):
                self.y[i] = np.argmin(np.linalg.norm(X[i] - self.mean_vec, axis=1))
            self.clusters = []
            for i in range(0, k):
                self.clusters.append(self.X[np.where(np.equal(self.y, i))])
            # Visualization
            if len(X[0]) == 2 and visual:
                for cluster in self.clusters:
                    plt.plot(cluster[:, 0], cluster[:, 1], 'o')
                    if len(cluster) > 2:
                        hull = ConvexHull(cluster)
                        plt.plot(cluster[hull.vertices, 0], cluster[hull.vertices, 1], 'r--', lw=2)
                        plt.plot(cluster[hull.vertices[[-1, 0]], 0], cluster[hull.vertices[[-1, 0]], 1], 'r--')
                plt.plot(self.mean_vec[:, 0], self.mean_vec[:, 1], 'r+')
                plt.xlabel('密度')
                plt.ylabel('含糖率')
                plt.title('第%d次迭代之后' % iter)
                plt.show()
            # Update mean vectors
            for i in range(0, k):
                next_vec = np.mean(self.clusters[i], axis=0)
                if np.linalg.norm(next_vec - self.mean_vec[i]) > epsilon:
                    self.mean_vec[i] = next_vec
                    stop = False
            iter += 1