On Google App Engine:
Google App Engine is Available for Java now - It was released earlier this
week. Check the link at
http://code.google.com/appengine/docs/java/gettingstarted/

My Motivation to ask for SDB is because of my use cases.

In my world - I have a constant stream of Preferences/Rating's flowing in
asynchronously. As a result in my basic implementation of CF,before I got to
Taste, I maintained a table for UserID to UserID Correlation score(Pearson
and Euclidean)  and ItemID to ItemID score - which were incrementally
updated on each  new  Userid - Item Id - Rating    transaction coming in.

I liked this model because the model was getting updated behind the scene's
constantly and for the consumer side(recommendations) I always had a userid
to userid score available to me for immediate access.

Given my usecase - I though that something like SDB would work fine because
my results were already in database(and did not need to be calculated
realtime) and I did not care about  ACID properties for database.

Would something like this work in TASTE.  I have still not gotten an
understanding of the implementation of Taste , but it seems that it
calculates the userid to userid similarity realtime.   Approaches like
dumping the preferences to S3 and then updating via hadoop don't gel in with
a Message Oriented Architectecture.

I will send the error-log in the next email



On Fri, Apr 10, 2009 at 2:34 PM, Sean Owen <[email protected]> wrote:

> On Fri, Apr 10, 2009 at 10:24 PM, Niraj <[email protected]>
> wrote:
> > I am using vista and apparently some tests are failing for
> >  mvn install for CORE.
>
> That's probably it -- can you provide any errors? it is almost surely
> something to do with paths, which I might be able to fix quickly. core
> and examples must build for this to work.
>
>
> > On another note: Has you out anythought to implementing TASTE for
> a)google
> > App Engine for Java and b) using simple db has the database
>
> App Engine is Python-only for the moment and no I would not seriously
> consider reimplementing in Python myself. But it is to support Java
> shortly. In that case, yes perhaps.
>
> I had already created a machine image (AMI) for Amazon's EC2 which
> would load data from S3, process all recommendations, and shut down.
> This sort of thing could be replicated on App Engine; it is a nice
> model.
>
> However, even better, Amazon just announced an easy framework for
> MapReduce (Hadoop) on EC2. On top of my to-do list is to put together
> the steps to run recommenders this way. The Hadoop bindings are
> already written; should be easy. This will be an even better model
> that the App Engine / EC2 model above. So I suppose I'd focus on this
> mostly.
>
> SimpleDB, Amazon's DB? I wouldn't object to a DataModel based on it.
> My concern is that recommenders are so data-intensive and access data
> so frequently that even a local database is quite slow -- let alone a
> distributed database like SimpleDB. I sense it is not cut out for this
> use case.
>

Reply via email to