Run a C++ model from python?

[jacksonsdean]

Is there a way to define a model in C++ and then run that model in a Python CUDA ensemble?

We have a model already created in C++, but we want to wrap an evolutionary algorithm around it in Python.

[Robadob]

Officially via FLAMEGPU, no. All our Python bindings are generated via SWIG. We tell it what we want wrapped in swig/python/flamegpu.i.

If you wanted to achieve this, you'd probably want to create a C++ method that returns (or fills) a ModelDescription with your model (I expect this would need to be part of the main flamegpu project. Then extend swig/python/flamegpu.i, so that SWIG additionally wraps that method to add it to the Python interface. If all is done correctly, when you rebuild pyflamegpu your method should now be available in the pyflamegpu module.

It's not something we've tried or required, and SWIG can be a bit unpredictable at times, so I'm not sure if any unexpected issues would occur whilst trying to get SWIG to play nicely.

[Robadob]

Theoretically, it should be possible to have SWIG wrap a C++ model in a separate (dependent) Python module. This would be a much cleaner solution that we could officially provide (via a template repository), however also more complicated to setup in the first place.

I've created an issue (#928) so that it's discussed by the team, but I don't expect it will be a priority.

[mondus]

Alternatively, translating the model to Python (using C++ style agent functions) should be fairly painless as there is a one to one mapping of all API calls. This obviously depends on if you have done anything with any other C++ libraries. In which case Robs answer is sensible. It would also be possible to have Python drive a C++ ensemble indirectly through file IO.