On Fri, Aug 22, 2014 at 11:52 PM, aman chourasia <[email protected]> wrote:
> Sebastian Schelter <ssc.open <at> googlemail.com> writes: > > > > > In Mahout 0.9, we will remove Slope One recommenders from the codebase. > > > > On 08.12.2013 16:03, Ted Dunning wrote: > > > Use a better recommender. Slope one is just there for completeness. > > > > > > > > > > > > Sent from my iPhone > > > > > >> On Dec 8, 2013, at 2:24, Siddharth Patnaik <spatnaik77 <at> gmail.com > > > wrote: > > >> > > >> What should be done to improve the runtime performance? > > > > > > can you please enlighten me with the reasons for removal of slope one > recommender, and what alternatives were included for this. > becuase as i see Slope One is very good for huge data which gives quite > good > result without any ovrefuitting. > That is interesting. All other feedback on slope-one was that it gave very poor results and simple cooccurrence approaches with LLR sparsification gives much better results at comparable or less computational cost. The original purpose of including it was just that it was simple enough to explain and had been mentioned in the literature. Eventually, that wasn't worth the problems that it caused and we removed it. We plan to remove more.
