A spark version about Factrorization Metric Learning for Ranking
mainly based on Metric Factorization: Recommendation beyond Matrix Factorization
codes structure:
src.jyb
FMModel.scala: main-model
package.scala: basic function
Param.scala: model-based parameters
test.scala: IO for model
Assumption:
the number of users >> the number of items
thus, distributed users in blocks (like mllib.ALS), and store items locally (as A map)
utilize AdaGard for adaptive adjusting study-rate