master branch for deployments @ICM
reverted to r52985 . Test runs shows we need to rely on the edgeIds produced by the connected components identfication phase instead of the vertexIds
simplified connected component application on the graph
do not skip processing datasets in DedupBuildRootsMapper, improved error reporting in DedupBuildRootsReducer
do not push vertex ids in memory, process them on the fly
beta
skip weird cases in CC algo
codebase used to migrate to java8 the production system