Please stay on the mailing list using reply-all if it's not the case by default.
2012/9/10 denis <[email protected]>: > Olivier, > ok, agree, but > http://scikit-learn.org/stable/auto_examples/cluster/plot_kmeans_digits.html#example-cluster-plot-kmeans-digits-py > suggests that KMeans on optdigits is reasonable, which it isn't; > how about adding your comment below to that page ? > also plot only mis-classified digits in color. It's not classification, it's clustering. We used a supervised signal only to evaluate the ability of the clustering algorithm to find some structure in the data that is approximately matching our high level human interpretation of it (the digits groups). ARI and AMI scores of more than 0.5 wrt some know ground truth is not that bad given the fact that the supervision signal (target classes) is not given to learning algorithm. But I agree this examples could be improved by comparing the performance with the results of SpectralClustering and adding some text to analyze the outcome. Pull Requests to improve examples are always welcome! If you want to do classification, then supervised models is strongly advised. Clustering is more useful for data exploration and maybe preprocessing for building new features (for instance a similarity value to the 3 or 5 closest centers of each samples). > (Bytheway to keep cluster sizes roughly the same as they grow, > find for each X the 3 nearest centres > then add X to the one with the best distance - λ clustersize). > > > Also a dumb question, can anyone tell me exactly how to subscribe to > scikit-learn-general > in Thunderbird ? Can't add [email protected] to Inbox ... I don't know about thunderbird, just follow the instructions of the mailing list manager here: https://lists.sourceforge.net/lists/listinfo/scikit-learn-general You don't need to fiddle with the thunderbird UI, just click on the confirmation email (maybe look in your spambox if it ended there). It just a regular mailing list, not a discussion group managed with the NNTP protocol. BTW, the canonical email address for the mailing list is [email protected], not [email protected]. -- 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
