Installing GeNN using requirements.txt on Streamlit

[russelljjarvis]

I am seeking to attach this adjustable network visualization of the Potjan's model to simulation parameters in the pygenn's Potjans model. Streamlit is in theory very easy to work with as a front end, however, you can only build things using a Python requirements.txt file and or apt-get commands.

Would it be viable to install pygenn on the streamlit server in such a way?

[neworderofjamie]

This should be possible by building your own wheel and installing via requirements.txt with a direct requirement - this should also work with URLs. In order to build a wheel you just need to run python setup.py bdist_wheel rather than python setup.py develop when building PyGeNN. The single-threaded CPU backend should work as long as the wheel is compatible. We are hoping to start distributing PyGeNN on pip but there are currently a few outstanding issues

[russelljjarvis]

I guess being a single CPU instance without NVIDIA GPU streamlit wouldn't leverage any of Genns strengths.

From this perspective, it might be worth renting a GPU enabled instance on DigitalOcean or AWS.

[neworderofjamie]

Yeah, the single-threaded CPU backend can be very competitive for small models (mostly because the code it generates is so simple compilers can do a pretty good job optimising it) but quickly gets slow for anything more than around 1000 neurons.