Investigate enabling training on gpu using Spark

[garlicbulbxian]

Purpose

We want to investigate to see if we can do the following things:

• Set up distributed Spark cluster with gpu instances
• Run training using elephas

  References

• http://christopher5106.github.io/big/data/2016/01/27/two-AMI-to-create-the-fastest-cluster-with-gpu-at-the-minimal-engineering-cost-with-EC2-NVIDIA-Spark-and-BIDMach.html