Usage of the 'block.cuba.Affine' layer

[ahenkes1]

Hello everybody,

I have a simple question with respect to the block.cuba.Affine layer. As far as I remember from a discussion with @bamsumit at the last workshop in Santa Clara, one way to readout the membrane potential of some neuron is using the respective Affine layer. As a simple example I use the following snippet:

import lava.lib.dl.slayer as slayer
import torch


class Network(torch.nn.Module):
    def __init__(self):
        """Initialize network."""
        super(Network, self).__init__()

        cuba_params = {
            "threshold": 0.1,
            "current_decay": 0.5,
            "voltage_decay": 0.5,
        }

        self.blocks = torch.nn.ModuleList(
            [
                slayer.block.cuba.Input(cuba_params),
                slayer.block.cuba.Dense(
                    cuba_params, in_neurons=1, out_neurons=1
                ),
                slayer.block.cuba.Affine(
                    cuba_params, in_neurons=1, out_neurons=1
                ),
            ]
        )

    def forward(self, x):
        for block in self.blocks:
            x = block(x)

        return x


def main():
    net = Network()

    x = torch.Tensor([0.1]).unsqueeze(-1).unsqueeze(-1).repeat((1, 1, 10))
    _ = net(x)

    return None


if __name__ == "__main__":
    main()

However, I get the following error:

File "/home/ah/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, **kwargs)
File "/home/ah/.local/lib/python3.10/site-packages/lava/lib/dl/slayer/block/base.py", line 339, in forward
x = self.neuron.dynamics(z)
File "/home/ah/.local/lib/python3.10/site-packages/lava/lib/dl/slayer/neuron/cuba.py", line 372, in dynamics
self.threshold + self.threshold_eps,
TypeError: unsupported operand type(s) for +: 'NoneType' and 'float'

Does anybody know what is the source of this error and how to fix it? I went through the documentation and the tutorials but could not find a fix. I the "bootstrap mnist" tutorial, a similar network topology is used, but it didn"t help me.

[ahenkes1]

As an update, setting the "dynamics" flag like

                slayer.block.cuba.Affine(
                    neuron_params=cuba_params,
                    in_neurons=1,
                    out_neurons=1,
                    dynamics=False,
                ),

seems to to the job.

Is this the correct way to achieve this? The documentation talks about dendrite current (Affine), I am not sure, if this is the same as the preceding layer potential.