* add NVML binding

* fix flake8

* separate gpu and nvml

* remove stuff

* switch to context

* Clean-up and rename

* fix lint

* Use tflayers

* some docs

* Use decorator for argument renaming

* rename parse_args

* Increase GPU counting robustness

Added native Tensorflow method to check number of GPUs.

* docs

* missing import

* fix import

Added dilation_rate to Conv2D since Tensorflow supports it easily. Requires TF 1.5 or greater.