Inconsistency about refractory period with Brian Simulator about bindsnet

Since there is a definition of t in network/init, one possible way is to add t as an additional argument to step function. Inside the step function, introduce new variable to store spiking time (say t_spike). Everytime step function is called, check if t-t_spike is greater than refractory period and do operation accordingly.

[djsaunde]

Good point. How would you fix this? Feel free to open a PR.

[djsaunde]

Hm, that could be problematic when doing multiple simulations in sequence. For example:

network.run(inpts, time=100)
network.run(inpts, time=100)  # Forgets refractory neurons from end of previous simulation!

Alternatively, one could just reverse the ordering, from:

# Decrement refractory counters.
self.refrac_count[self.refrac_count != 0] -= dt 

# Integrate inputs. 
self.v += (self.refrac_count == 0).float() * inpts 
  
# Check for spiking neurons.
self.s = self.v >= self.thresh

# Refractoriness and voltage reset. 
self.refrac_count.masked_fill_(self.s, self.refrac) 

to

# Integrate inputs. 
self.v += (self.refrac_count == 0).float() * inpts 
  
# Decrement refractory counters.
self.refrac_count[self.refrac_count != 0] -= dt 

# Check for spiking neurons.
self.s = self.v >= self.thresh

# Refractoriness and voltage reset. 
self.refrac_count.masked_fill_(self.s, self.refrac) 

Does this make sense? Does this appear to solve the problem?

Yes, that will be better than my solution.

You may need to be careful when changing order. There are models (like this) that do self.refrac_count == 0 more than once.

[djsaunde]

Solved by #136.