I am running the CIFAR-10 Classifier with a Spiking CNN notebook in Nengo Extras 0.3, but I run into this error on line 6. What does this mean and how do I fix it?
First of all, be warned that this is a very old example. NengoExtras hasn’t been kept up-to-date with the rest of the Nengo ecosystem, and we have newer/better ways of implementing convolutional networks. You might want to check out one of these examples:
- Optimizing a spiking neural network — NengoDL 3.4.1.dev0 docs (this is an MNIST example, but wouldn’t be too hard to do CIFAR-10 instead)
- CIFAR-10 convolutional network — NengoLoihi 1.1.0.dev0 docs (this is a CIFAR-10 example for NengoLoihi, but most of the notebook is actually in NengoDL for training the network. If you’re really set on using NengoOCL, that network is created in pure Nengo, which you could initialize with your trained weights, and then run in NengoOCL. Note that for running in NengoDL you will need to use the patch here: Fixed functools.partial fix in Cifar10 notebook by xchoo · Pull Request #315 · nengo/nengo-loihi · GitHub).
As for your specific error, it’s caused by a mismatch between your Nengo and NengoOCL versions, specifically that your Nengo version is too new for your NengoOCL version. The latest release of NengoOCL (v2.1.0) is compatible with Nengo 3.1.0. See here for more info about the NengoOCL releases and which Nengo versions they’re compatible with.