MNIST Spinnaker Example about snn_toolbox

[Aashish3970]

Hello,
I tried using the MNIST keras spinnaker example https://github.com/NeuromorphicProcessorProject/snn_toolbox/blob/master/examples/mnist_keras_spiNNaker.py. When I deploy this on spinnaker, the moving accuracy of ANN is 100% the moving accuracy of SNN is 0%. After that, I changed the simulation duration to 150 and v_thresh to 0.75. I am still getting the same accuracy. Plus, I get Average spike rate :0 spikes per simulation time step. What am I doing wrong?

Below is my config file

[rbodo]

Hi, I'm afraid I'll have to refer you with this issue to one of the SpiNNaker developers who were involved with this example (ej159, pabogdan). I currently don't have access to the hardware and cannot test this script myself. Sorry for not being able to help with this.

[Aashish3970]

Thanks for the reply. Does the toolbox create the PyNN code of the SNN and save it somewhere? How can I extract this SNN converted code ?

[rbodo]

The toolbox saves the pyNN projections here:

snn_toolbox/snntoolbox/simulation/target_simulators/spiNNaker_target_sim.py

Line 313 in 2a64a9d

def save_connections(self, path):

There will be a txt file for each layer containing its connections.

If you use the pyNN backend in the toolbox rather than SpiNNaker, it will also save the pyNN assemblies etc:

snn_toolbox/snntoolbox/simulation/target_simulators/pyNN_target_sim.py

Line 235 in 2a64a9d

def save(self, path, filename):

Maybe also take a look at the load_assembly method in that file, which reconstructs the layers from the saved files.

[Aashish3970]

Thanks a lot.