Hi Mohit, an addition to my previous e-mail: I am currently implementing a case where users and items do have some differences in their value. So what I did, I created an extra field in the vertex value which is only used when the vertex represents an item. When the vertex represents a user, this field remains unused.
If you - or anybody else - has a different approach, don't feel shy to share it with us :) Cheers, On Thu, Jul 18, 2013 at 3:36 PM, Maria Stylianou <[email protected]> wrote: > Hey Mohit, > > Having vertices of different type does not actually affect you. > A vertex has four attributes: <Id, value, edge, message> > Talking for the Collaborative Filtering algorithms, both users and items > will be characterized similarly in terms of a vertex; each user/item needs > an Id to be unique, has its own value, a set of edges that carry data, and > a set of messages that transfer data. In practice you will notice that both > users and items carry data of the same type. > > Cheers, > > > On Wed, Jul 17, 2013 at 8:31 PM, Mohit Singh <[email protected]> wrote: > >> >> Hi, >> I am new to giraph. I have gone thru some of the examples (very basic >> ones). >> I have one question. How can I implement algorithms like LDA or >> collaborative filtering in giraph. >> The reason I ask this is, usually such algorithms are implemented by a >> bipartite graph (user-item graph or doc-word graph). >> How do i implement algorithms whose vertices are of different type. >> Thanks >> >> -- >> Mohit >> >> "When you want success as badly as you want the air, then you will get >> it. There is no other secret of success." >> -Socrates > > > > > -- > Maria Stylianou > Intern at Telefonica, Barcelona, Spain > Master Student of European Master in Distributed > Computing<http://www.kth.se/en/studies/programmes/master/em/emdc> > marsty5.wordpress.com > > -- Maria Stylianou Intern at Telefonica, Barcelona, Spain Master Student of European Master in Distributed Computing<http://www.kth.se/en/studies/programmes/master/em/emdc> marsty5.wordpress.com
