Is there a 'value range' of INITIAL CENTROIDS for kmeans and kmeans ++?

Hi,

We know that the initial centroids are randomly generated for Matlab's default 'kmeans' function.

I wonder whether these initial centroids are randomly generated in a 'specific range', say in the range of [0, 1], or there is no such range and the value of initial centroids could be any number.

I also have the same question for 'kmeans ++' about the value range of the initial centroids, if there is such 'value range'.

Are negative values allowed for initial centroids??

Answer 1

[the cyclist]

The "kmeans++" algorithm is the algorithm for choosing the initial centroids for the kmeans procedure. You can see that in the documentation here. So, I don't think you are really asking two separate questions.

The initial centroids are not chosen "from a range", nor are they "any number". The initial centroids are chosen from the input data itself. The first initial centroid is one of the data points, selected at random. After that, there is a probabilistic algorithm (based on the distance from other initial centroids) for choosing the next one.

It is all spelled out in the documentation.