BIDData/BIDMach
We recently ran some fresh benchmarks for Spark v1.1 and v1.2 and Graphlab clusters, and included some updated numbers from other recent published benchmarks. RCV1-v2 (Reuters news data, LYR2004 distribution) benchmarks are for OAA (One Against All) classification, since RCV1-v2 has 103 independent topic labels. RCV1-v2 is a small dataset (0.5 GB). BIDMach was run on a single Amazon g2.xlarge instance, while Spark was run on a cluster of m3.2xlarge high-memory instances. The other systems were run on an 8-code Intel E-2660 system.
Apr-3-2016, 19:45:39 GMT