It starts with the distance graph. That alone is not very scalable. For the special case of Euclidean distances between unit vectors you can use dot product instead which is preserved approximately with the svd you can make it a little better but you still have a quadratic algorithm.
Sent from my iPad On Oct 5, 2011, at 8:50 PM, Lance Norskog <[email protected]> wrote: > http://dspace.mit.edu/handle/1721.1/17671 > > Is this usable? Is the mixture model trick feasible for mahout's target > scale? > > -- > Lance Norskog > [email protected]
