FAFU机器学习09-2 Clustering Methods课件.pptxVIP

Foundations of Machine LearningClustering Methods （聚类方法）2023/11/4Clustering MethodsLesson 9 - 1 Clustering MethodsPartitional Methods （基于划分的方法），prototype-based clustering （基于原型的聚类）：K-meansHierarchical Clustering （层次聚类）Density-based Clustering （密度聚类）2023/11/4Clustering MethodsLesson 9 - 2 K-means algorithms （k均值算法）Center-basedA cluster is a set of objects such that an object in a cluster is closer (more similar) to the “center” of a cluster, than to the center of any other clusterThe center of a cluster is called centroidEach point is assigned to the cluster with the closest centroidThe number of clusters usually should be specified2023/11/4Clustering MethodsLesson 9 - 3 K-means algorithms基本步骤Partition {x1,…, xn} into K clusters （K is predefined）InitializationSpecify the initial cluster centers (centroids)Iteration until no changeFor each object xi (Cluster Assignment )Calculate the distances between xi and the K centroids(Re)assign xi to the cluster whose centroid is the closest to xiUpdate the cluster centroids based on current assignment (Update Cluster Centroid)2023/11/4Clustering MethodsLesson 9 - 4 K-means: Initialization2023/11/4Clustering MethodsLesson 9 - 5 K-means Clustering: Cluster Assignment2023/11/4Clustering MethodsLesson 9 - 6 K-means Clustering: Update Cluster Centroid2023/11/4Clustering MethodsLesson 9 - 7 Sum of Squared Error (SSE)K-means 算法的优化目标：误差平方和Suppose the centroid of cluster Ci is ui For each object x in Ci, compute the squared error between x and the centroid uiSum up the error of all the objects2023/11/4Clustering MethodsLesson 9 - 8 K-means 算法伪代码2023/11/4Clustering MethodsLesson 9 - 9 Comments on the K-Means MethodStrengthEfficient: O(tkn), where n is # objects, k is # clusters, and t is # iterations. Normally, k, t nEasy to implementIssuesNeed to specify K, the number of clustersLocal minimum– Initialization mattersEmpty clusters may appear2023/11/4Clustering MethodsLesson 9 - 10 Pre-processing and Post-processingPre-processingNorm