Open rgerkin opened 6 years ago
Is this needed in light of recent PyNN -> NeuroML2 support (see here)? Maybe the model class could just implement this conversion and then run through the LEMSModel
class.
Duplicate of #27
I guess it all depends on how seamlessly that conversion process works.
My goal is to enable sciunit tests, of network behavior via SpiNNaker and PyNN. I have an intuition that writing a PyNN interface (a psuedo backend), is still the right way to go.
I was also thinking, given that PyNN models, are almost implemented models that have neo Analogue signal outuputs, why not create a Model Class: ExternalModel, as follows:
https://github.com/russelljjarvis/neuronunit/blob/dev/neuronunit/models/__init__.py#L24-L41
The purpose of this ExternalModel, is to avoid reimplementing the mundane (get_membrane_potential etc), and in fact instead of over-riding these methods with inheritance, over ride these methods, by assigning the external models own similar methods, to get_membrane_potential
etc.
PyNN support should come via a model class
PyNNModel
that wraps PyNN, rather than a backend. PyNN will supply its own means for communicating with backend simulators. ThePyNNModel
should be analogous toLEMSModel
, and will be a parallel approach to instantiating a SciUnit model for testing.