I need to find a good clustering for this data using sci-kit.
KNN is not appropriate as it creates blobs although these data are linearly separated.
import hypertools as hyp
hyp.plot(tocluster,'o', n_clusters=10)
$\begingroup$
$\endgroup$
1
asked Jun 3, 2017 at 9:09
-
$\begingroup$ This page is a good reference for pros and cons of different clustering algos. Maybe gaussian mixture model might be a good starting place for this data scikit-learn.org/stable/auto_examples/cluster/… $\endgroup$– Eric A. BunchCommented Dec 6, 2018 at 19:36
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
I'd rather not rely on clustering (clearly, DBSCAN would be the first method to try).
Instead, I'd look for a projection that removes the correlation (x - y maybe?), then define a grid to separate the data.
answered Jun 3, 2017 at 11:37