• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    迪恩网络公众号

xuyxu/Clustering: Clustering / Subspace Clustering Algorithms on MATLAB

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

xuyxu/Clustering

开源软件地址(OpenSource Url):

https://github.com/xuyxu/Clustering

开源编程语言(OpenSource Language):

MATLAB 100.0%

开源软件介绍(OpenSource Introduction):

Clustering/Subspace Clustering Algorithms on MATLAB

This repo is no longer in active development. However, any problem on implementations of existing algorithms is welcomed. [Oct, 2020]

1. Clustering Algorithms

  • K-means
  • K-means++
    • Generally speaking, this algorithm is similar to K-means;
    • Unlike classic K-means randomly choosing initial centroids, a better initialization procedure is integrated into K-means++, where observations far from existing centroids have higher probabilities of being chosen as the next centroid.
    • The initializeation procedure can be achieved using Fitness Proportionate Selection.
  • ISODATA (Iterative Self-Organizing Data Analysis)
    • To be brief, ISODATA introduces two additional operations: Splitting and Merging;
    • When the number of observations within one class is less than one pre-defined threshold, ISODATA merges two classes with minimum between-class distance;
    • When the within-class variance of one class exceeds one pre-defined threshold, ISODATA splits this class into two different sub-classes.
  • Mean Shift
    • For each point x, find neighbors, calculate mean vector m, update x = m, until x == m;
    • Non-parametric model, no need to specify the number of classes;
    • No structure priori.
  • DBSCAN (Density-Based Spatial Clustering of Application with Noise)
    • Starting with pre-selected core objects, DBSCAN extends each cluster based on the connectivity between data points;
    • DBSCAN takes noisy data into consideration, hence robust to outliers;
    • Choosing good parameters can be hard without prior knowledge;
  • Gaussian Mixture Model (GMM)
  • LVQ (Learning Vector Quantization)

2. Subspace Clustering Algorithms

  • Subspace K-means
    • This algorithm directly extends K-means to Subspace Clustering through multiplying each dimension dj by one weight mj (s.t. sum(mj)=1, j=1,2,...,p);
    • It can be efficiently sovled in an Expectation-Maximization (EM) fashion. In each E-step, it updates weights, centroids using Lagrange Multiplier;
    • This rough algorithm suffers from the problem on its favor of using just a few dimensions when clustering sparse data;
  • Entropy-Weighting Subspace K-means
    • Generally speaking, this algorithm is similar to Subspace K-means;
    • In addition, it introduces one regularization item related to weight entropy into the objective function, in order to mitigate the aforementioned problem in Subspace K-means.
    • Apart from its succinctness and efficiency, it works well on a broad range of real-world datasets.



鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap