Nicer model initialisation

[picciama]

Need to discuss how we handle this: https://github.com/theislab/batchglm/blob/604823001e81fdab232fe27dc59dab8d506ea020/batchglm/models/base_glm/model.py#L66-L82

There are basically three options in my opinion:

1. Remove InputDataGLM entirely: We only use it as a container now that does some type checks in the beginning which we could do statically in utils anyway.
2. Keep InputDataGLM and reference the attributes in the model properties, i.e. model.input_data.<attribute> when calling model.attribute
3. override model.__getattr__ like so:

   def __getattr__(self, attr: str):
       return self.input_data.__getattribute__(attr)
   

I'm in favour of 2 but as a way in between to keep this support, maybe 3 would be an elegant solution for now.

Please advise @ilan-gold @davidsebfischer

[ilan-gold]

What does 2 allow that 3 doesn't? Not sure I follow the what you mean by "keep this support"

[picciama]

Sorry I meant 1. 2 would probably be cleaner than 3 but other than that not really a different as in both casese input_data is a model attribute