@Gael
> > > I believe that it is the same thing as cosine similarity. If that's
> > > indeed the case, you could add a note in the cosine similarity
docstring
> > > to stress it.
> > I think it is somewhat different from cosine similarity.
> Then you'll have to tell me how, because I am being dense and I don't see
> the difference.
Sorry, I should've been specific :|
> > > I remember there is an off-the-shelf function in scipy.stats called
> > > pearsonr. You don't have to implement it on your own.
> > Yeah, I know about that. I thought of suggesting this addition after I
> > saw that we *have* a newton_cg as comparted to scipy's fmin_ncg. :)
> Our newton_cg actually has a different implementation than scipy's
> fmin_ncg and these differences are necessary to make the logistic
> regression significantly faster.
I didn't mean to say that the two functions have the same implementation.
Was just highlighting the fact that we could use one too, perhaps with a
different implementation or some new options.
@Boyuan
> Hi Vinayak:
> scipy.stats implemented pearsonr() like that because it's a statistics
> routine. It treats 0 in the input data as indeed value 0.
> But in the context of recommender systems, "unrated" is different from
> score 0 (though we usually use 0 to represent "unrated" when score must
> be positive). And strictly speaking, Pearson correlation here is
> *defined* on only the items that both users have rated.
> You can use routines like numpy.nonzero() to get the indices of commonly
> rated items before calling pearsonr() in scipy.stats.
> In my opinion, providing such new options for pearsonr() will confuse
> users who are not doing recommender systems and make it deviate from its
> initial definition.
I agree that those options could confuse users and make it deviate from its
initial definition.
@All - How should we go about it?
Vinayak
------------------------------------------------------------------------------
Dive into the World of Parallel Programming The Go Parallel Website, sponsored
by Intel and developed in partnership with Slashdot Media, is your hub for all
things parallel software development, from weekly thought leadership blogs to
news, videos, case studies, tutorials and more. Take a look and join the
conversation now. http://goparallel.sourceforge.net/
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
