Stochastic SVD and alternating least squares would both be better, I think.
On Tue, Jul 31, 2012 at 7:28 PM, Aniruddha Basak <[email protected]>wrote: > I did not choose Lanczos. It found it being used as a part of > Spectral-Kmeans > implemented in Mahout. > Can u please recommend some better alternative of Lanczos. > > Thanks, > Aniruddha > > > -----Original Message----- > From: Ted Dunning [mailto:[email protected]] > Sent: Tuesday, July 31, 2012 6:24 PM > To: [email protected] > Cc: Jake Mannix > Subject: Re: Mahout LanczosSolver explanation > > WHy are you using Lanczos? Why not use something more recent? > > On Tue, Jul 31, 2012 at 7:00 PM, Aniruddha Basak <[email protected] > >wrote: > > > Hi, > > I am working on Spectral Kmeans which involves an eigen-decomposition > > step using Lanczos. As I did not get exact similar results as > > expected, I tried to understand the implementation. > > I have one question about "LanczosSolver .java" : > > In the "solve" method, while building the "tridiagonal matrix" there > > is a step just after the multiplication job (performed on Hadoop as > > TimesSquaredJob) as > > ------------ > > nextVector.assign(new Scale(1.0 / state.getScaleFactor())); > > ----------- > > I could not understand why this scaling is performed? > > > > ( When I compared the results on a small matrix to an equivalent > > Matlab script, I found the results are exactly similar WITHOUT this > > scaling. Including this scaling makes the results different from the > > Matlab results. ) > > > > > > Thanks, > > Aniruddha > > > > >
