2014-04-11 10:55 GMT+02:00 Daniel Vainsencher <[email protected]>:
> In any case, the approximate nature of the search raises the possibility
> of going a step further: index the data points, and adjust each cluster
> to its ANNs (in this case, for a very long list of candidates). This is
> no longer k-means (closer to a mean-shift algorithm) and may or may not
> work, but could be very fast.

Speaking of, mean-shift is already implemented using NN. Judging from
GitHub issues, ML questions and the complexity notes in the mean-shift
docstrings, I also believe that optimizing it would be more valuable
than optimizing k-means, since we already have minibatch k-means.

(Also k-means can still benefit from the Elkan optimization, which
doesn't change its semantics.)

------------------------------------------------------------------------------
Put Bad Developers to Shame
Dominate Development with Jenkins Continuous Integration
Continuously Automate Build, Test & Deployment 
Start a new project now. Try Jenkins in the cloud.
http://p.sf.net/sfu/13600_Cloudbees
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to