It doesn't have to do with the mean preference value being 0. The problem is that these users overlap in only one item, and so, Pearson cannot compute a similarity since it cannot define a correlation coefficient. A correlation would require at least two data points. Add some more data and you will get non-NaN values.
On Tue, Jun 23, 2009 at 6:38 PM, charlysf<[email protected]> wrote: > > Hello, > > I use PearsonCorrelation, and I would like to understand why on my example, > I have a NaN. > http://www.nabble.com/file/p24175732/Image%2B3.png > > http://www.nabble.com/file/p24175732/Image%2B4.png > > > I use PearsonCorrelation for an Item Basis recommender, and I of course > expect, to have a similarity between the two items, because they are linked > to a same user. > > But in fact, I have a NaN. > > I saw that the mean of the data should be 0, but or me, 0 is the neutral > score, and I can't be sure that for each user, their mean is 0, is it > important ? I don't really understand why. > > > Thank you. > -- > View this message in context: > http://www.nabble.com/Questions-about-PearsonCorrelation-on-a-example-tp24175732p24175732.html > Sent from the Mahout User List mailing list archive at Nabble.com. >
