I am trying to use Nengo in order to convert a ANN → SNN and then running this SNN on a SpiNNaker machine.
In order to convert the ANN → SNN I need the Nengo_DL package, while the nengo_spinnaker package takes care of running the SNN on the spinnaker machine.
Nengo_DL and Nengo_SpiNNaker seem to be incompatible since Nengo_DL requires Nengo version >= 3.0.0 while Nengo_Spinnaker requires Nengo version 2.2.0.
Hmmm. It will be a little tricky, considering that the NengoSpinnaker package is quite old and not actively maintained. You can attempt to solve this conflict by installing the following package versions: nengo==2.2.0 nengo_dl==2.1.1 tensorflow==1.15
I’m not 100% sure this will work though, as there may be unforeseen complications, especially with the TensorFlow version.
I do know that Nvidia maintains their own TensorFlow 1.15 builds that work better with modern GPUs.
I tried installing the specific package versions but I noticed that the respective versions you mentioned are not compatible. Installing the versions you mentioned raises a ValueError from the Nengo_DL version.py .
Instead I used the following packages and there are no conflicts:
I’m pretty sure TensorFlow 2.5.0 would not work with NengoDL 1.0.0 since NengoDL 1.0.0 was written before TF 2 was even conceived. But, you can still give it a shot to see if it works.