mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: RecommenderJob and GenericUserBasedRecommender
Date Fri, 21 Aug 2015 19:14:49 GMT
The hadoop recommenders are not the really state of the art anymore. 

We are recommending one based on the SimilarityAnalysis.cooccurrence methods and the spark-itemsimilarity
driver if you wish. These in combination with a search engine produce an incredibly flexible
recommender that ingests lots more data about the user and so can recommend in cases the old
ones or even the new Spark ALS recommender can’t help with.

There is an OSS version here: https://github.com/PredictionIO/template-scala-parallel-universal-recommendation
It works as a template or plugin to the prediction.io framework, which has all the event ingest/storage/train/serving
built in. Check it out.

On Aug 20, 2015, at 12:30 PM, Mike C <mike@siteswithbenefits.com> wrote:

Hi!

I've made a custom User Recommender using the Mahout API, using
GenericUserBasedRecommender.

I'm not quite sure how to take the next step to get it working across
Hadoop.  RecommenderJob appears to be for Item Recommenders and it's not
really clear how to adapt it for a custom Recommender.

Anyone have any pointers to how this can be done in my case?

Thanks.

Mike


Mime
View raw message