Cluster shape in agglomerative clustering #19436
Replies: 2 comments
-
I am not sure how to quantify the shape of an n-dimensional cluster. Do you have pointers in the literature for such "shape metrics"? Maybe plotting the histograms of distances to the centroid could be informative? I am pretty sure we have nothing like this in scikit-learn. |
Beta Was this translation helpful? Give feedback.
-
If you want to "calculate the shape of clusters", I would recommend having a look at some topological data analysis (TDA) libraries like Gudhi. Given clusters represented as numpy arrays, Gudhi can compute various simplicial complexes. From those complexes, you can compute persistence diagrams which quantify topological features of your clusters. If TDA is new to you, I would recommend reading [1]. /cc @mglisse, @VincentRouvreau and @MathieuCarriere which have way more authority than me on this subject. [1]: Frédéric Chazal, Bertrand Michel. An introduction to Topological Data Analysis: fundamental and practical aspects for data scientists. 2017. hal-01614384 |
Beta Was this translation helpful? Give feedback.
-
Hi,
Is there already a way to calculate the shape of the clusters after the data points are assigned cluster values?
Eg : spherical, rod-like, ribbon-like ,spiral etc.
Beta Was this translation helpful? Give feedback.
All reactions