Paper and code could benefit from rewrite

[mattdornfeld]

Both the arxiv paper and the code in this repo for the deep kalman filter algorithm are somewhat unclear. In the paper it is rather difficult to understand the exact algorithmic steps used to train the network. I believe clarification of this issue will lead to wider use of the algorithm by the community.