Kmeans clustering mahalanobis distance

Can anyone share the code of kmeans clustering that uses mahalanobis distance instead of euclidean distance

Answer 1

[the cyclist]

Will kmedoids do what you want?

[shaifali gupta]

Kmedoids also has linear boundary.. because it again uses euclidean.. So guess no

[the cyclist]

The documentation link I sent states that one can specify mahalanobis as the distance. (See the section on Name-Value arguments.)

Full disclosure: I have not used this function myself, so maybe it is not as simple as it seems to me.

[shaifali gupta]

Sir.. which documentation link are you referring to? I checked here and its does not mahalanobis distance in its parameters. https://in.mathworks.com/help/stats/kmeans.html

[the cyclist]

There is a link in my answer. Note that it is a link to the kmedoids function, not the kmeans function.

[shaifali gupta]

I am trying it... It gives error The covariance matrix for the Mahalanobis metric must be symmetric and positive definite.

Can you tell how can I resolve it

[the cyclist]

It seems that your current question has been answered here. (Maybe someone else will else also add some other answers, of course.)

I suggest you open a new question now, for specific help on working with the kmedoids function. In that new question, please add more detail, perhaps adding a MAT file with your data, or a small example of what is not working for you. As I mentioned, I am not very familiar with that function, and you might get help from more people with a new question.

Answer 2

[白]

You could try nearestPSD, which could transform your covariance matrix into symmetric and positive definite one, but there is some estimation bias to check.