Conv1D Probe with LinearFilter half the size of output

Getting Started
1

Hi I have implemented a large neural network in Tensorflow and converted it to Nengo with nengo_dl Converter,
however when I add Probes to the Conv1D with bias layers, they become mismatched when using nengo.LinearFilter([1], [1, 0]) and Nengo throw an error during simulation.
The Conv1D layers seem double the size, I figure it is because of the biases, but cannot find where the size of the synaptic filter is decided.

Due to the network being so large I cannot keep the history with going out of memory, and wanted to use the integrator to count spikes.

2

Hi Ceasar,

if you can post a minimal example script here i’ll try to figure out what’s going on!

3

Hi thank you for the quick reply!

I couldn’t hold it, so I figure it out.

Seemed to be an issue with loading network parameters.
I could not add Probes with LinearFilters before loading the parameters, despite the environment being the same.

I solved it by changing the Probes synapses inside the simulation environment after loading the network parameters.

4

Great!