This is a fix for issue #136. #138
Conversation
… for situations where sub-populations of neuron populations were connected and a state monitor for the synapses instantiated, the pre-template prep code would identify the size of the connected sub-populations for the srcN and trgN variables of the stateMonitorModel. These variables are then used for the argument list of the convert_dense_matrix_2_dynamic_arrays and convert_sparse_synapses_2_dynamic_arrays calls that translate GeNN synaptic data into brian format synaptic data. Because brian2genn implements sub-populations only by connectivity matrices between full populations, the conversion functions need the size of the full populations, not of the connected sub-populations. I believe the srcN and trgN variables are solely for teh purpose of these conversions, so should not be needed to represent subpopulation size elsewhere.
|
Oh dear ... I see the tests are failing but I am afraid it is likely due to a change in Brian 2. The tests seem to pass locally with brian 2.4.2 |
|
Oops, based on the b2g release notes, I assumed compatibility with later brian2 versions was not guaranteed anyway, and didn't report my own failures on later/current brian2. Which is to say: Can confirm, brian2 > 2.4.2 breaks brian2genn. |
|
Apologies, Brian2GeNN is indeed not compatible yet with Brian 2.5, but requirements and test suite do not reflect this. I'll try to take care of this tomorrow. |
[ci skip]
|
I pushed changes to master to make Brian2GeNN compatible with Brian 2.5 and merged the changes into this branch. All tests pass, so this is good to merge. |
|
Excellent, thank you both! |
The root cause for the problem was that for situations where sub-populations of neuron populations were connected
and a state monitor for the synapses instantiated, the pre-template prep code would identify the size of the connected sub-populations for the
srcNandtrgNvariables of thestateMonitorModel. These variables are then used for the argument list of theconvert_dense_matrix_2_dynamic_arraysandconvert_sparse_synapses_2_dynamic_arrayscalls that translate GeNN synaptic data into brian format synaptic data. Because brian2genn implements sub-populations only by connectivity matrices between full populations, the conversion functions need the size of the full populations, not of the connected sub-populations.I believe the
srcNandtrgNvariables are solely for the purpose of these conversions, so should not be needed to represent sub-population size elsewhere.