Retrieve run-time threshold condition

Support
#1

Hi folks, I am experimenting with implementing the non-Markovian Gillespie Algorithm (nMGA) in Brian based on Vestergaard and Genois (PLoS Comput. Biol. 2015, p9). Is there a way to retrieve real-time evaluations of the threshold/reset conditions?

At the moment my method is implemented by reaction=NeuronGroup(1,code=nMGA,...) where

nMGA = '''
       tau : 1
       Q   : 1
       transition = tau<Q*dt : boolean
       
       ... 
       ...
       
       '''

Say I have a reaction in the nMGA scheme every time tau<Q*dt where tau and Q are set inside my nMGA implementation (within reaction=NeuronGroup(...). This reaction is detected passing threshold='tau<Q*dt to the NeuronGroup.

Because the nMGA requires some tricks for multiple transitions within the same time step, I am using reaction.run_regularly(code=code_after_transition) as well, where

code_after_transition = '''
                        if not transition:
                        ...
                        else
                        ...
                        '''

I want to avoid defining transition in the reaction equations and have it instead directly passed from the reaction.thresholder or whatever evaluates the threshold condition. Something in the spirit of lastspike.

#2

Hi Maurizio,
we don’t have a variable that is explicitly set by the thresholder, spikes are stored in a dedicated data structure. I have to admit that I do not exactly understand the issue: why do you not want to define transtion in the equations? If you wanted to set a variable after each spike (or transition), you could set it in the reset. Or the same for a custom event and run_on_event.

#3

@mstimberg You are totally right. Sorry. I was caught in a classic loop-of-thoughts. To answer your question. The idea is that I have a transition = tau<Q*dt : boolean in the equations. And then if (not transition): <do some stuff> else: <do some other stuff upon transition>. This is the case of the nMGA, with the complication, that upon a transition, the algorithm requires to do some additional stuff after the reset and before updating to the next time step.

On a side note, following up on to your reply, do the reset or threshold code strings allow specification of complex code in the form of if (...) : else : ... or for or while loops (the latter for example in the reset code)? Or do I need to resort run_on_event in these cases?

#4

You cannot use control structures like if/else or for/while in Brian code, the only direct way to implement it would be in C++/Cython code as part of user-defined functions. If you want to assign two different values to a variable depending on a condition, you can use the workaround var = int(condition)*value_1 + int(not condition)*value_2. Otherwise you’ll need to do everything with custom events and run_on_event, e.g. you can do stuff like:

group = NeuronGroup(..., events={'transitioned': 'transition',
                                 'no_transition': 'not transition'})
group.run_on_event('transitioned', 'do something')
group.run_on_event('no_transition', 'do something else')
1 Like