... might be done within the Graphinius IO project instead of the core library
- at least test the iML API with some demo / virtual user which provides same / different answers each time
- then automatically run the original algorithm again on the dataset, anonymizing it with the newly learned weight vector