Implement the pairwise metrics

[marcelcaraciolo]

It will be implemented the pairwise metrics with new signatures in order to get the same code for several functions.

For instance consider the pearson correlation which can be expressed as cosine pairwise metric changing the inputs. (Follow the link: http://pt.wikipedia.org/wiki/Coeficiente_de_correla%C3%A7%C3%A3o_de_Pearson)

The mahantan and euclidean metrics can be considered a Lp-norm changing the p = {1,2,3 ...}

Before commiting I recommend you test without optimizations to check those assertives and post the gist so we can disccuss.

[irgmedeiros]

More on distances optimization can be found here: http://en.wikipedia.org/wiki/Minkowski_distance and here: http://en.wikipedia.org/wiki/Lp_space

[irgmedeiros]

@marcelcaraciolo I implemented the adjusted cosine similarity using the cosine similarity under the hood. The same idea can be applied to others similarities. https://gist.github.com/irgmedeiros/5859643. The doc test example uses the example of the book http://guidetodatamining.com/guide/ch3/DataMining-ch3.pdf page 17. Take a look.