Lets say you have an algorithm in which a vector starts out sparse, but gradually fills in. At some point, you want to give up and switch to an ordinary vector. This came up in Brown and DiPietro's entropy-based bigram information word clustering algorithm, and I think it could come up elsewhere.
A more complex class based on what you're doing here would use this value to decide when to give up and use a dense vector instead. On Mon, Jun 15, 2009 at 7:42 PM, Sean Owen<sro...@gmail.com> wrote: > I don't dispute it but what is the use case? I am mostly curious at this > point. > > On Tue, Jun 16, 2009 at 12:40 AM, Benson Margulies<bimargul...@gmail.com> > wrote: >> The 'numberOfNonDefaultValueElements' is useful. I'd give it an >> accessor, with, well, that very name. >> >> On Mon, Jun 15, 2009 at 5:20 PM, Sean Owen<sro...@gmail.com> wrote: >>> OK don't want to push on this last bit too much, but I still see a >>> small concern here. >
