Re: [Pcsim-users] Question about ConnectionsProjection

[Mathieu D. <mat...@li...>]

Hi Dejan,

Thanks for the answer.
>   
>> Hello PCSIMers,
>>
>> Sorry to bother you again but I was wondering why ConnectionsProjection 
>> doesn't inherit from SimObjectPopulation?
>>
>> I ask the question because I often want to create recorders for a whole 
>> ConnectionsProjection (for instance to record the psr). With a 
>> SimObjectPopulation, it's easy: the record method is here for that. But 
>> AFAIK there is no equivalent method in ConnectionsProjection. So I often 
>> have to create a population from a ConnectionsProjection with something 
>> like syn_popul = SimObjectPopulation( proj.idVector() ).
>>   
>>     
> The is not any strong reason why ConnectionsProjection does not
> inherit from SimObjectPopulation. There is a difference in the
> behavior of ConnectionsProjection in distributed modes of simulation
> that does not conform to the definition of the SimObjectPopulation,
> like for example, it's possible to specify collectIDs = false,
> in the ConnectionsProjection which saves memory by not storing
> the ids of the synapses. Also in some distributed construction algorithms,
> a ConnectionsProjection instance on one MPI node stores the IDs of the 
> synapses
> that are located on the particular node, whereas a SimObjectPopulation 
> in distributed
> mode is expected to hold the all IDs of SimObjects over all nodes.
> Nevertheless, in spite of these differences which can be explicitly 
> pointed out in the documentation
> in order to avoid confusion and reconciled with error messages in the 
> interface if the IDs are not present
> in the projection, it should be in principle possible to make 
> ConnectionsProjection inherit from SimObjectPopulation,
> and we have considered it. It's not done since other things had higher 
> priority. 
> For recording variables from synapses I also create a 
> SimObjectPopulation from the ID vector of synapses.
> It's possible that the interface is extended with this feature in the 
> future.
>   
OK. I'll continue to use the SimObjectPopulation trick (I have probably 
borrowed it from an example :).
I'll be happy to test this feature in a future release.

>> Also why ConnectionsProjection doesn't store information about the 
>> source and destination populations? Therefore we could write something 
>> like proj.getSourcePopulation (which again could be useful in my case).
>>   
>>     
> This is also a possibility that is not yet implemented. Usually in a 
> script when you create a ConnectionsProjection,
> you can keep the python references to both source and dest. populations 
> that you can use afterwards, therefore we
> didn't consider it important or necessary to add it so far.
> However, I have noted this, and we will consider to change this in the 
> future.
>   
OK. One reason for that is that I have written some simple functions to 
get some properties of the network. For instance a function to list 
which neurons (identified by their local indexes in the population) are 
connected by a ConnectionsProjection (see attachment).
As the ConnectionsProjection itself doesn't store information about the 
populations it connects I have to pass those populations as arguments.
As an aside this could probably be done via NeuroML but I have never 
used it.
> Thank you for your suggestions.
>   
I have seen a "Feature Request" on the SourceForge page. Would that help 
if I submit those suggestions here?

Best regards,
Mathieu