-1

I did some clustering with sklearn's Affinity Propagation and now I want to check how my clustering performs. I know there are a lot of metrics to check that, but some questions have appeared:

  1. Which kind of metrics to use? MSE, EMSE? (I was clustering trajectories/paths of moving objects)
  2. Metrics usually require true values and predicted values. But the thing is that I did not predict anything. I just clustered given data. What to fix to predicted or to true data?
Mamed
  • 101
  • Possible duplicate of [How to select a clustering method? How to validate a cluster solution (to warrant the method choice)?](https://stats.stackexchange.com/questions/195456/how-to-select-a-clustering-method-how-to-validate-a-cluster-solution-to-warran) – Sycorax Sep 04 '19 at 00:43

1 Answers1

1

The right choice of metric obviously depends a lot on your task. So your question cannot be answered because we don't know what you are trying to do... There is no "one size fits all" way, or it would automatically be computed for you.

There are plenty of evaluation measures - both internal and external - in sklearn! So you'll need to study them to identify which one you want to use for your goal. It's probably not MSE for trajectories.

Clustering is not classification. But in some sense you do predict. You only use a different vocabulary - you predict a cluster number (at least in a flat clustering as AP). The exact value does not matter, but it matters which objects have the same cluster and which are in different clusters. If you have at least partially labeled data, then you can use this kind of evaluation.

Has QUIT--Anony-Mousse
  • 39,639
  • 7
  • 61
  • 96