Create a mechanism to interrupt a simulation

[mstimberg]

This could be enabled automatically via an except hook, or be an option to Network.run. It would basically mean that pressing Ctrl+C during a run would call Network.stop(), so that the network would be in a well-defined state after the interruption.

[thesamovar]

The excepthook version could be a Brian preference (I'd leave it off by default)?

[Jasmine969]

Hi Brian2 developers, is there any progress on this issue? In my work, I want to halt a simulation if a certain quantity is out of range. But there seems no method in Brian2 that can do this.

[mstimberg]

Hi Brian2 developers, is there any progress on this issue? In my work, I want to halt a simulation if a certain quantity is out of range. But there seems no method in Brian2 that can do this.

@Jasmine969 Apologies, I've completely overlooked this question (or maybe you asked it on the forum as well)? This issue is now closed since we added the Ctrl+C mechanism. For more general stopping as you seem to have in mind, you can do this with a network_operation in runtime mode: see https://brian2.readthedocs.io/en/latest/examples/network_operation_stop.html Unfortunately, we don't have a corresponding mechanism in place for C++ standalone. However, with the recent change you could manually write a user-provided function with C++ code, and set brian::Network::globally_stopped = true; when your condition is fulfilled.

[Jasmine969]

Thanks. But network_operation will slow down the simulation significantly.

[mstimberg]

Thanks. But network_operation will slow down the simulation significantly.

Well, it depends – in this type of use case, you probably do not have to do the check every timestep? If you provide a large dt, the effect should be negligible. Also, one has to be careful to do the check in the most performant way, e.g. rather use the unitless version of a value, do not loop over arrays, etc.