### KMeans++ 实现与应用 KMeans++ 是一种改进版的 K-Means 聚类算法，通过优化初始质心的选择来提高收敛速度和最终模型的质量。该算法在数据科学领域广泛应用，在 Python 中可以通过 `scikit-learn` 库轻松实现。 #### 使用 Scikit-Learn 进行 KMeans++ Scikit-Learn 提供了一个内置支持 KMeans++ 的接口，允许用户快速构建并训练聚类模型[^1]： ```python from sklearn.cluster import KMeans import numpy as np # 创建随机样本数据集 X = np.random.rand(100, 2) # 初始化带有 k-means++ 参数的 KMeans 对象 kmeans_plusplus = KMeans(n_clusters=3, init='k-means++') # 训练模型 kmeans_plusplus.fit(X) # 输出分类标签 print(kmeans_plusplus.labels_) ``` 此代码片段展示了如何利用 scikit-learn 来执行基于 KMeans++ 的聚类分析。值得注意的是，默认情况下，`init='k-means++'` 已经被设置为 True，因此无需额外指定即可享受其带来的性能提升效果。 #### 手动实现 KMeans++ 对于希望深入了解工作原理的学习者来说，下面是一个简单的手动实现版本: ```python def initialize_centroids(points, k): """初始化质心""" centroids = points.sample(n=k).values.tolist() while len(set(tuple(c) for c in centroids)) != k: centroids = points.sample(n=k).values.tolist() return centroids def closest_centroid(point, centroids): """计算最近距离的质心""" distances = [np.linalg.norm(np.array(point)-np.array(cen)) for cen in centroids] return np.argmin(distances) def move_centroids(points, assignments, k): """移动质心到簇中心位置""" new_centroids = [] for i in range(k): assigned_points = [points[j] for j in range(len(points)) if assignments[j]==i] centroid = np.mean(assigned_points, axis=0) new_centroids.append(list(centroid)) return new_centroids def k_means_pp(dataframe, num_clusters, max_iterations=300): """完整的 KMeans++ 算法流程""" # Step 1: 随机选取第一个质心 initial_point = dataframe.iloc[np.random.randint(0,len(dataframe))] centroids = [initial_point.values] # Step 2: 根据概率分布选择剩余质心 for _ in range(num_clusters - 1): D_x = [] for point_index in range(len(dataframe)): min_distance = float('inf') for center in centroids: dist = np.linalg.norm(center-dataframe.iloc[point_index].values)**2 if dist < min_distance: min_distance = dist D_x.append(min_distance) sum_D_x = sum(D_x) probabilities = [d/sum_D_x for d in D_x] cumulative_probabilities = np.cumsum(probabilities) r = np.random.uniform() selected_center_idx = next(i for i,p in enumerate(cumulative_probabilities) if p >= r) centroids.append(dataframe.iloc[selected_center_idx].values) previous_assignments = None # 开始迭代更新过程... for iteration in range(max_iterations): assignments = [] for index,row in dataframe.iterrows(): assignments.append(closest_centroid(row,centroids)) if assignments == previous_assignments: break previous_assignments = assignments centroids = move_centroids(dataframe.values,assignments,num_clusters) return centroids, assignments ``` 这段自定义函数实现了整个 KMeans++ 流程，包括但不限于初始化阶段的概率加权采样以及后续的标准 K-Means 更新逻辑。