mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sebastian Schelter <...@apache.org>
Subject Re: Deploying a massively scalable recommender system with Apache Mahout
Date Thu, 04 Aug 2011 10:13:15 GMT
On 04.08.2011 12:07, David Cabanillas wrote:
> Hello,
>
> On Thu, Aug 4, 2011 at 12:04 PM, Sebastian Schelter <ssc@apache.org
> <mailto:ssc@apache.org>> wrote:
>
>     On 04.08.2011 11:58, David Cabanillas wrote:
>
>         Sorry Sebastien and I have only test the cf working with mysql,
>         but I
>         have not any example.
>
>         Nowadays I am in an initial state, my idea is to apply
>         similarity in a
>         job portal, the idea is to recommend to unemployed skills that other
>         unemployed have selected. For example, if many users knows Java and
>         Pascal and a new user selects Java the system should recommend
>         Pascal too.
>
>
>     How many datapoints will the system have to process for this? I'm
>     not sure you really need to use hadoop for that. An in-memory
>     recommender might be a much easier to deploy solution.
>
>
> Problably we have 1000 - 2000 users with 20-40 items per user. Do you
> know any in-memory recommender resource?

That's definitely nothing you need hadoop for. The easiest way to tackle 
your usecase would be to get a copy of "Mahout in Action" and read the 
recommendation chapter. This covers everything you need for your usecase.

If you don't want to spend any money, read those websites:

https://cwiki.apache.org/confluence/display/MAHOUT/Recommender+Documentation

http://blog.jteam.nl/2009/12/09/mahout-taste-part-one-introduction/

--sebastian

>
>
>     --sebastian
>
>
>         Thanks for your support.
>         PS: At the end If I would have done the solution (to use mahout with
>         mysql) I have published this solution in your blog.
>
>
>         On Thu, Aug 4, 2011 at 11:48 AM, Sebastian Schelter
>         <ssc@apache.org <mailto:ssc@apache.org>
>         <mailto:ssc@apache.org <mailto:ssc@apache.org>>> wrote:
>
>             David,
>
>             it is not helpful and frustrating, if you don't answer
>         questions.
>
>             I think you are on a wrong track currently, to quote my
>         blogpost:
>         "Be aware that this is a guide intended for readers already familiar
>             with Collaborative Filtering and recommender systems that are
>             evaluating Mahout as a choice for building their production
>         systems
>             on. The focus is on making the right engineering decisions
>         rather
>             than on explaining algorithms here."
>
>
>             And please reply to the user-mailinglist and not to me in
>         person,
>             the purpose of Apache projects offering support is to have
>         public
>             conversations and give all readers the possibility to learn
>         not to
>             have free private consultation by the committers.
>
>
>             --sebastian
>
>
>             On 04.08.2011 11:44, David Cabanillas wrote:
>
>                 Right now I only want to connect mahout with mysql and I
>         have
>                 not find
>                 any example.
>                 In the section *Putting the puzzle together you said:
>                 *
>
>                 DataSource datasource = ...
>
>
>                 But what's means ... ???
>
>
>                 On Thu, Aug 4, 2011 at 10:14 AM, Sebastian Schelter
>         <ssc@apache.org <mailto:ssc@apache.org> <mailto:ssc@apache.org
>         <mailto:ssc@apache.org>>
>         <mailto:ssc@apache.org <mailto:ssc@apache.org>
>         <mailto:ssc@apache.org <mailto:ssc@apache.org>>>> wrote:
>
>                     David, can you please give us some details about
>         your usecase?
>
>                     It seems like you're trying to reimplement the system I
>                 described in
>         http://ssc.io/deploying-a-______massively-scalable-______recommender-system-with-______apache-mahout/
>         <http://ssc.io/deploying-a-____massively-scalable-____recommender-system-with-____apache-mahout/>
>         <http://ssc.io/deploying-a-____massively-scalable-____recommender-system-with-____apache-mahout/
>         <http://ssc.io/deploying-a-__massively-scalable-__recommender-system-with-__apache-mahout/>>
>         <http://ssc.io/deploying-a-____massively-scalable-____recommender-system-with-____apache-mahout/
>         <http://ssc.io/deploying-a-__massively-scalable-__recommender-system-with-__apache-mahout/>
>         <http://ssc.io/deploying-a-__massively-scalable-__recommender-system-with-__apache-mahout/
>         <http://ssc.io/deploying-a-massively-scalable-recommender-system-with-apache-mahout/>>>
>
>                     That system is highly optimized for a certain class of
>                 usecases and
>                     only makes sense if you have like 100+ million
>         datapoints
>                 and 100+
>                     requests/second to your recommender.
>
>                     If you just want to start diving into recommendation
>         mining and
>                     build a first system to play with, working with this
>         article is
>                     definitely the wrong approach. In that case, I highly
>                 suggest you
>                     get a copy of "Mahout in Action",
>         http://manning.com/owen/ which
>                     gives a superb introduction to recommendation mining
>         with
>                 mahout.
>
>                     --sebastian
>
>
>                     On 03.08.2011 14:59, David Cabanillas wrote:
>
>                         Hello Sebastian,
>
>                         Right now, I have the precomputed item-similarity my
>                 problem is to
>                         relate it with mysql.
>                         In section*Setting up the infrastructure for the
>         live
>                 recommender
>                         system* you suggest that we should to use
>                 MySQLJDBCDataModel, tu
>                         I don't
>                         understand how it works.
>
>                         Don't have any code example to relate mahout and
>         mysql?
>                         Many thanks.
>
>                         bye
>                         --david
>
>
>
>
>
>                 --
>                 bye
>                 --david
>
>
>
>
>
>         --
>         bye
>         --david
>
>
>
>
>
> --
> bye
> --david


Mime
View raw message