Total tangent: Anyone doing scientific computing work on the JVM may be interested in:

ND4J: N-dimensional arrays for the JVM http://nd4j.org/

Libnd4j: The C++ engine powering the above https://github.com/deeplearning4j/libnd4j

JavaCPP: The bridge between Java and C++ (Cython for Java) https://github.com/bytedeco/javacpp

Fwiw, all that works on Spark with multi-GPUs.

http://deeplearning4j.org/spark

http://deeplearning4j.org/gpu