Closed pgleeson closed 1 year ago
@vvbragin can you please look into this? maybe related to some of the latest changes? thx
Hi @pgleeson, thanks for sharing this. I've found out that it is the case only for transition from 1.0.2.1 to later versions, and only for delays of stimSource-stimTargetParams. We had a minor issue in stims generation code, one side effect of which was that it used to re-pick values from random distribution more than once for each stim (so the original value was being overridden by subsequent ones). After we fixed it in 1.0.3, it generates correctly, but this also means the values are different from now on.
@salvadord CC
Ok, thanks for this clarification @vvbragin. I'll update my tests to use the latest values and assume those will be generated from this version on.
Using the same seed for network generation should produce the same set of connections/weights/delays across different versions of NetPyNE presumably.
This script HHSmall.py generates a NetPyNE net and exports it to NeuroML, and has consistently been producing the same connections, but now is producing different delays:
The first set of delays are the same, but those in the 2nd projection are different (all values come directly from what's generated by NetPyNE)