0
点赞
收藏
分享

微信扫一扫

聚类分析①

zhongjh 2022-01-26 阅读 101

(方法名称)Parameters(参数)Scalability(可扩展性)Usecase(使用场景)Geometry (metric used)(几何图形(公制使用))
K-Means(K-均值)number of clusters(聚类形成的簇的个数)非常大的 n_samples, 中等的 n_clusters 使用 MiniBatch 代码)通用, 均匀的 cluster size(簇大小), flat geometry(平面几何), 不是太多的 clusters(簇)Distances between points(点之间的距离)
Affinity propagationdamping(阻尼), sample preference(样本偏好)Not scalable with n_samples(n_samples 不可扩展)Many clusters, uneven cluster size, non-flat geometry(许多簇,不均匀的簇大小,非平面几何)Graph distance (e.g. nearest-neighbor graph)(图距离(例如,最近邻图))
Mean-shiftbandwidth(带宽)Not scalable with n_samples (n_samples不可扩展)Many clusters, uneven cluster size, non-flat geometry(许多簇,不均匀的簇大小,非平面几何)Distances between points(点之间的距离)
Spectral clusteringnumber of clusters(簇的个数)中等的 n_samples, 小的 n_clustersFew clusters, even cluster size, non-flat geometry(几个簇,均匀的簇大小,非平面几何)Graph distance (e.g. nearest-neighbor graph)(图距离(例如最近邻图))
Ward hierarchical clusteringnumber of clusters(簇的个数)大的 n_samples 和 n_clustersMany clusters, possibly connectivity constraints(很多的簇,可能连接限制)Distances between points(点之间的距离)
Agglomerative clusteringnumber of clusters(簇的个数), linkage type(链接类型), distance(距离)大的 n_samples 和 n_clustersMany clusters, possibly connectivity constraints, non Euclidean distances(很多簇,可能连接限制,非欧氏距离)Any pairwise distance(任意成对距离)
DBSCANneighborhood size(neighborhood 的大小)非常大的 n_samples, 中等的 n_clustersNon-flat geometry, uneven cluster sizes(非平面几何,不均匀的簇大小)Distances between nearest points(最近点之间的距离)
Gaussian mixtures(高斯混合)many(很多)Not scalable(不可扩展)Flat geometry, good for density estimation(平面几何,适用于密度估计)Mahalanobis distances to centers( 与中心的马氏距离)
Birchbranching factor(分支因子), threshold(阈值), optional global clusterer(可选全局簇).大的 n_clusters 和 n_samplesLarge dataset, outlier removal, data reduction.(大型数据集,异常值去除,数据简化)Euclidean distance between points(点之间的欧氏距离)
举报

相关推荐

0 条评论