Getting never more than 3 clusters, unless manual initial points were given.

[jonathanslenders]

Somehow, we get never more than 3 clusters. Except for a few points, all of them fall into a single cluster.

We are using postgis, using the following query:

SELECT id, kmeans(ARRAY[ ST_X(point), ST_Y(point)], 40) OVER () AS k, id, FROM our_table ORDER BY k DESC;

Is there any way to improve the initial choosen points?

[umitanuki]

Sorry for the delay. Picking up initial points automatically is a hard problem, but I thought I did something reasonable... Could you provide reproducible dataset? I think I could add more various methods for the initial points like kmeans++ if there are needs.