2012/5/25 Gael Varoquaux <[email protected]>: > On Fri, May 25, 2012 at 09:43:29AM +0200, bthirion wrote: >> > labels = np.unique(labels, return_index=True)[1][labels] >> -0 >> Why not, but this is easy and safe to do only in some cases: >> -- do not forget to permute all the label-related info (cluster centers, >> weights, covariance)... >> -- In case of hierarchical clustering, you need to decide whether you >> break the consistency of the labelling across level of the hierarchy. > > I agree. I was thinking of doing it only for a small number of clustering > algorithms. I had in mind in particular kmeans. What gave me this idea > was that testing kmeans was harder than it should.
Indeed the cluster centers will have to be updated to. I am not sure I really see the use case: is it to make sure that cluster plots use the same colors? Do you have code that is dependent on the exact cluster label indices? I think I am neither +1 nor -1. A concrete use case (example?) would help further motivate this change. -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
