2
$\begingroup$

I am trying to identify rows in groups of points using clustering algorithms. The bigger picture problem I'm trying to solve is to identify shelves given x and y coordinates of products. I can cluster based on just the y-coordinates and I get decent clustering using algorithms such as HDBSCAN.

I am wondering if it is possible to also incorporate the x-coordinates somehow into the clustering as it feels like a waste of helpful information to simply discard the x-coordinated.

However, if I put the x-coordinates into my clustering algorithm (i.e. each point is now represented by a 2D vector), the clustering is poor because two points at opposite ends on the x-axis are considered not related as compared to two points directly below each other albeit on different shelves.

enter image description here

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ If you know the number of shelves before-hand (and shelves are level) I do think k-means on just the y is a good approach. Looking at the x coord residuals within a cluster may help to see that you have an entire shelf covered though (they should be uniform and not have holes). $\endgroup$
    – Andy W
    Commented Dec 4 at 13:45

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.