Assigned clusters to new data

K-means clustering as such is an unsupervised method. From what I gather, you would like to learn the cluster centroids using the kmeans algorithm and then use these centroids to map new test data to the centroids in some manner. You could do the following :-

1. Assign each new data point to its closest centroid, by using a distance measure like sum-of-squares(same as Euclidean distance), or cosine similarity. For these, you can simply use the formulae, or use built in methods such as norm (also refer: How to calculate Euclidean distance ) or pdist.

Another (perhaps better) way, instead of reusing the centroids :-

2. Label all the samples in your training set according to the cluster they were assigned to (ex. you can choose a cluster and label all the points inside it as belonging to class 1), and then train a classifier (could be any algorithm, such as SVM) on this training data. Afterwards, classify your test samples using this model.

Hope this helps!

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

Answer 2

KSSV 2018-6-1

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/403517-assigned-clusters-to-new-data#answer_322774

Read about knnsearch. This gives you the nearest neighbors of the given point. If your neighbors falls in certain Id..the given point also falls in the same ID.