The current setup.py install_requires= argument includes graphviz as a dependency.

This should be factored out as a required dependency, similar to how scikit-learn does not require this package, but it's needed for some model visualization directives.

Pull Request #6 includes the ability for models to be saved and loaded from json files (rfgb.rdn.learn). Currently there are no unit tests for this functionality though.

At a minimum, a unit test might check whether the files are created. Then they might also test whether the contents of the json files are the same as the parameters input during inference.