Hi, The matrix if sparse can be very large like 1000 X 1000 but it will only have at most 20 non-zero elements. That is the reason I specifically talked in terms of DenseMatrix since a sparse matrix or Vector would waste a lot of space and traversal time.
Distance comparison can be simply by subtracting one matrix from another and the elements of resulting matrix should be squared and added. On Wed, Jun 20, 2012 at 5:38 PM, gaurav singh <[email protected]>wrote: > Hi, > > The matrix if sparse can be very large like 1000 X 1000 but it will only > have at most 20 non-zero elements. That is the reason I specifically talked > in terms of DenseMatrix since a sparse matrix or Vector would waste a lot > of space and traversal time. > > > > On Wed, Jun 20, 2012 at 4:34 PM, Ted Dunning <[email protected]>wrote: > >> Yeah... you can probably do this. It will involve storing your matrices >> as >> vectors and probably requires that they be the same size. >> >> Can you say more about the matrices in terms of size and how you compute >> distance? >> >> On Wed, Jun 20, 2012 at 1:42 AM, gaurav singh <[email protected] >> >wrote: >> >> > Hi all, >> > >> > I can see that kmeans can be used to create clusters of vectors but can >> we >> > somehow use kmeans of mahout to create clusters of matrices. In other >> words >> > do we already have distance calculation functionality between matrices >> in >> > mahout and does kmeans implementation support that? >> > >> > I wish to be able to convert certain sequences into markov model and >> > generate 2d arrays. I would like to cluster these sequences on the >> basis of >> > the distance between different markov matrices. I would wish them to be >> > dense matrices and I also know that we have hadoop writable DenseMatrix >> in >> > class in mahout. >> > >> > I just wish to know if it can be directly used with kmeans or I will >> have >> > to write or customize kmeans for my purpose? >> > >> > Thanks for any help offered! >> > >> > -- >> > Regards >> > Gaurav Singh >> > >> > > > > -- > Regards > Gaurav Singh > > -- Regards Gaurav Singh
