tag:blogger.com,1999:blog-3211409948956809184.post4464223586215709102..comments2024-03-21T04:14:27.443-07:00Comments on Large Scale Machine Learning and Other Animals: Incremental SVDDanny Bicksonhttp://www.blogger.com/profile/01517237836051035400noreply@blogger.comBlogger3125tag:blogger.com,1999:blog-3211409948956809184.post-51259312135003240182013-05-03T16:56:01.316-07:002013-05-03T16:56:01.316-07:00There is work by Gu and Eisenstat on incremental S...There is work by Gu and Eisenstat on incremental SVD.<br /><br />Now regarding his problem:<br /><br />1) He is really interested in computing the SVD of A, so no need to form A'*A (which is actually not the suggested way to compute the SVD of A). One can use dense SVD methods, and compute all singular values. I would suggest trying the package Elemental.<br /><br />2) Since he wants only a few hundred singular values, iterative methods might work better. I would suggest taking a look at ARPACK (it computes eigenvalues, but you can compute the eigenvalus of [0 A'; A 0]).<br /><br />3) If low accuracy is enough, one can try the randomized PCA methods. Just google those.Anonymoushttps://www.blogger.com/profile/13979542076247424819noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-28759785834383828652013-04-29T23:10:42.704-07:002013-04-29T23:10:42.704-07:00I don't know him in person - do you know him?
...I don't know him in person - do you know him?<br />:-)Danny Bicksonhttps://www.blogger.com/profile/01517237836051035400noreply@blogger.comtag:blogger.com,1999:blog-3211409948956809184.post-73795066618606415152013-04-29T16:12:10.216-07:002013-04-29T16:12:10.216-07:00Or ask Matthew directly ?Or ask Matthew directly ?Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.com