Bad default values for on_target and off_target in mse_membrane_loss

[saiftyfirst]

mse_membrane_loss uses spikegen to convert a tensor of targets into a tensor of spikes. For example:

target = torch.tensor([2, 1, 0]) 
on_target = 1
off_target = 0
targets_spikes = spikegen.targets_convert(
        targets, num_classes=3 , on_target=on_target, off_target=off_target)

The problem
mse_membrane_loss defines member variables on_target and off_target by default with values of 0 and -1. This results in a target_spike tensor of all 0s (Try above example with the changed on_target and off_target values)

Code Reference / Cause
spikegen.py:1023
function: targets_rate(...)

if on_target > off_target:
                return torch.clamp(
                    to_one_hot(targets, num_classes) * on_target, off_target
                )