I think that just getting good hash functions is plenty good enough. On Mon, Apr 25, 2011 at 6:37 PM, Dmitriy Lyubimov <[email protected]> wrote:
> if you are talking about cosine similarity distance lsh family, would > it help to do the sphere projection first? will actually also decrease > the dimensionality by one if i am not mistaken > > -d > > On Mon, Apr 25, 2011 at 4:32 PM, Ted Dunning <[email protected]> > wrote: > > Yes. Mean subtraction would do the trick. > > > > On Mon, Apr 25, 2011 at 4:20 PM, Jake Mannix <[email protected]> > wrote: > > > >> > It may be that in practice that SVD will scatter data into many > orthants, > >> > but I suspect it will not spread the data as widely as LSH assumptions > >> > would > >> > like. > >> > >> > >> Maybe you're right - raw SVD vs PCA (where the means are also subtracted > >> off) is probably the distinction to draw here. > > >
