we loaded the index to memory, it takes <3s per request for 2M users.
On Thu, Jan 27, 2011 at 8:42 AM, Sean Owen <[email protected]> wrote: > You'll want to make sure you're pooling (but sounds like that's all set) > and > the indexes are in place. > > But, it'll never perform well querying from a database. You want to load it > into memory. Look at the ReloadFromJDBCDataModel. > > On Thu, Jan 27, 2011 at 4:24 PM, Saul Moncada <[email protected]> > wrote: > > > Hi all, > > > > I'm using the GenericItemBasedRecommender (Mahout 0.5 Snapshot) in Tomcat > 6 > > with PostgreSQL 9 as database backend and a C3PO connection pool (50 > > concurrent connections). > > > > Since my items are ordered in a tree I'm using the Wu & Palmer > correlation > > to calculate the item similarity. > > > > My dataset contains 17000 items, 279000 similarities and 5000 user > > preferences. > > > > My current problem is that it takes about 15 seconds to give one > > recommendation to a user. > > > > Based on your experience and the dataset size (which should grow a lot in > > the next few months) do you think this time is normal? I'd like to use > the > > recommender in real time. Should I change it to precalculate the > > recommendations offline? > > > > Regards, > > > > Saúl Moncada > > > > -- > > Visita: http://www.kentriki.com.ve > > Compras - Vendes - Compartes > > >
