How do I evaluate a recommender using Mahout/Hadoop -
mahout comes number of predefined hadoop job classes (stuff derived abstractjob). these include jobs computing recommendations (recommenderjob). mahout defines various recommenderevaluators use in testing these recommendations, it's not clear me how 1 runs these in distributed fashion using hadoop.
is it: run recommender job, , feed output of evaluator? looked @ rmsrecommenderevaluatortest, seems everythinig in 1 fell swoop, , doesn't hadoop job.
firstly think recommendations in mahout come in 2 parts: non-distributed algorithms (from taste) , distributed hadoop jobs (parallelals , item-based collaborative filtering). rmseevaluator in non-distributed part, don't think works hadoop jobs.
for evaluating hadoop recommendation job in mahout, have @
org.apache.mahout.cf.taste.hadoop.als.factorizationevaluator
you can write scripts , download recommendation results local filesystem evaluation.
Comments
Post a Comment