tag:blogger.com,1999:blog-3211409948956809184.post6822912024287972433..comments2024-03-21T04:14:27.443-07:00Comments on Large Scale Machine Learning and Other Animals: Power iteration clusteringDanny Bicksonhttp://www.blogger.com/profile/01517237836051035400noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-3211409948956809184.post-46448620858013914432012-02-17T08:02:02.581-08:002012-02-17T08:02:02.581-08:00Did you ever get this implemented?
Sounds like it...Did you ever get this implemented?<br /><br />Sounds like it might also be a natural fit with pregel-like APIs e.g. Giraph. Sound plaisible?Danhttps://www.blogger.com/profile/00785005036504488460noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-75918134215417391512011-12-10T12:42:22.019-08:002011-12-10T12:42:22.019-08:00Hi!
This is the answer i got from Frank Lin: In ge...Hi!<br />This is the answer i got from Frank Lin: In general, PIC should work well where spectral clustering (spec. NCut) works well. With more clusters (perhaps usps) more than one pic vectors should be used.<br /><br />One caveat using these methods in general is how to do the similarity embedding, so that's my guess; but I'd like to try them and see why they don't work so well, if indeed they don't.Danny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-82250247981377176712011-12-10T05:25:05.394-08:002011-12-10T05:25:05.394-08:00based on my experiments, the performance on wine, ...based on my experiments, the performance on wine, usps and sona data sets are not good.cswhjianghttp://nonoreply@blogger.com