Clustering Azimuths: Algorithm & Link

[billiards]

Hi all,

If I have a set of azimuths, e.g. [ 0, 10, 11, 67, 68, 69, 70, 124, 127, 136, 355].

How can I cluster these directions bearing in mind that 355 is close to 0?

Can someone point me to a link, preferably with an algorithm I can use.

Cheers

 

[mfb]

I'm sure there are algorithms for clustering on a (1-dimensional) torus. A quick google search pointed me to this.
If you don't find one, this hack might work: transform your one-dimensional distribution to a circle in 2 dimensions: x -> (sin x, cos x), and look for clusters there.

 

[billiards]

Nice idea to use sinx, cosx.

Any idea what the best clustering algorithm to use would be? Ideally I want something that can figure out the optimum number of clusters itself from the data.

(Incidentally I am trying to do this using python -- so any python specific help would be particularly appreciated)

 

[mfb]

No idea, sorry.

 

[CellCoree]

Hello,

Thank you for your question. Clustering azimuths is a common problem in many scientific fields, including geology, astronomy, and meteorology. There are several approaches that can be used to cluster these directions, depending on the specific application and data set.

One approach is to use a circular statistics method, such as the circular k-means algorithm. This method takes into account the circular nature of azimuths and can handle cases where 355 is close to 0. The algorithm can be implemented using various programming languages, such as R or Python, and there are many online resources that provide step-by-step instructions and code examples.

You may also find this paper helpful: "Circular K-means clustering for directional data" by L. Batschelet (1981). It provides a detailed explanation of the algorithm and its applications.

Additionally, there are software packages such as Oriana and OrianaJ that have built-in circular k-means clustering functions and can handle large data sets.

I hope this helps. Best of luck with your research!