Monitoring Model During Simulation

[a2l5t8]

When I simulate my network, I would like to know how much time and space is used for each component of my network in order to help optimize it, Is there any way to save the amount of time it takes for each behavior or each object like NeuronGroups to simulate without manually calculating these values?

[realamirhe]

Here's a code snippet I use to identify and troubleshoot bottlenecks:

from pyinstrument import Profiler

with Profiler(async_mode="disabled") as p:
  run_your_conex_model_here()
  
p.print()

[a2l5t8]

in run_your_conex_model_here() do need to implement all the network definition or is it okay to only include only the following?
net.initialize() net.simulate_iterations(ITERATIONS)

[realamirhe]

Cause the net.simulate_iterations(ITERATIONS) is the most time-consuming part, via its iterative nature, it suffices to only run network simulation within Profile with-context

[saeedark]

Pymonntorch's Network class has a method for measuring the blocks' time.

https://github.com/cnrl/PymoNNtorch/blob/0f8243dac802fb340a558f2c1d2618ab5cbf7fb9/pymonntorch/NetworkCore/Network.py#L290

(Turn measure_behavior_execution_time to True for a single iteration.)
Due to the asynchronous nature of GPU execution, the measured times are not accurate, and therefore using a profiler is advised.