Put forward and inverse models in explauto/models ?
Closed this issue · 4 comments
Should we put explauto/sensorimotor_model/forward and inverse in explauto/models ?
Pro: models should be in /models
Con: forward and inverse models are only used in sensorimotor_model
Hi Sébastien,
The original idea was rather to have the algorithms in /models (in particular if they used elsewhere that in sensorimotor_model, i.e for the interest models) and the interfaces (i.e. subclasses of SensorimotorModel) in /sensorimotor_models. This is for example what was done for the IMLE model.
I'm actually not really aware of the fact that we now have two separate subdirectories, forward and inverse. It seems a bit strange to me since most models can be used both in a forward and in an inverse way. Do you know what was the reason of this split?
The forward/inverse split comes from the adapted import of the useful part of Fabien's code, that was previously hidden in "third_party". This allow in particular to easily combine any forward model with any inverse model algorithm (e.g. LWLR-BFGS, LWLR-CMAES).
Currently only NN and WNN appear in both, but the algorithms for those forward and inverse models are not the same, specially for WNN.
IMLE and ILO-GMM, are directly in /sensorimotor_models as they compute both forward and inverse models.
OK. So to reply to your previous question:
I would put in explauto/models the code that can potentially be used in different packages (i.e. not only in sensoriotor_model).
Or more practically speaking: I would address this issue only at the time when it becomes problematic for whatever reason. Is it the case now?
Ok. No problematic reasons, it was just to know what is the best way.
So yes let's keep that as I think nothing will be used elsewhere.
The only code that is used also in interest_model is the dataset.py so I already put it in /models.