Simplified a little bit and removed private parameters, but thats pretty much the code. We did not try with toy examples, since that was already done with the example pipelines when we implemented and the model training itself is quite simple basic there already (only few hyperparameters set)