I am trying to implement Converting a Keras model to an SNN on Loihi with a different dataset. I try to keep everything same like MNIST. First I try to test my network with MNIST dataset and it works completely fine. But when i try to replace the MNIST dataset it shows an error “‘function’ object has no attribute ‘shape’ in nengo” in train the network with normal ReLU neurons sections. My train dataset shape is (10000, 28, 28) and test shape is (800, 28, 28).
Could you please suggest me how to fix the issue?
I am having a similar issue. I was trying to have my custom dataset working with the sample code of “Optimizing a spiking neural network”. As you described, the sample code works just fine with the MNIST dataset. I get the following (returning NaN) whenever I substitute MNIST dataset with my custom data set. The custom data set has the same format with MNIST data. It is just of different single channel 28x28 images.