Adding new methods to the Apache mahout

Question

I am trying to do some work and research on recommender systems. I want to use Apache Mahout. However, I want to add some methods as there is no implementations for them now. I tried to do that but the source code of Mahout is locked and I cannot add or change anything. I'm wondering if anyone knows how to resolve this issue.

Answer 1

Fork the project on Gihub into your own repository. Make the modifications you want and if they are good enough create a "pull request" back to Mahout. They may be incorporated into the mainstream.

Be aware that Mahout does not accept old style Hadoop "mapreduce" code. The new Mahout is based on more modern compute backends, mostly Spark but more recently Flink. The project is also primarily Scala although this is not a requirement.

The modern Mahout Recommender support is provided through SimilarityAnalysis.cooccurrence and SimilarityAnalysis.rowsimilarity . These provide for collaborative filtering and content-based recommendations but must be served using a knn/search engine. See Mahout docs here and a full running recommender here: https://templates.prediction.io/PredictionIO/template-scala-parallel-universal-recommendation