issue with time-varying edge variables when multiple nodes are used

[pyrates-neuroscience]

issue with time-varying edge variables when multiple nodes are used

[NinaOmejc]

Hi,

the setting vectorize=False worked and solve my issue, thank you very much!

[Richert]

Hi Nina,

Thank you for bringing this to my attention! There indeed seems to be a bug when using edge operators with multiple input variables in combination with the vectorize=True keyword argument of the CircuitTemplate.run method (which is also the default value of that keyword argument).

I will look into this and make sure that this bug is fixed as soon as possible. Until then, you should be able to use vectorize=False as a temporary bugfix.

Best, Richard

[NinaOmejc]

I have one more question, somehow related to this, if you are willing to discuss.

Currently, my time varying variable of coupling strength is inside the node operator and is thus affecting the coupling strength to all other nodes inside a network. Is there a way to add a time varying parameter to the edge operator? I read that the "weight" can only be a scalar. Also, when I tried to add an "input variable" to the "equation" and "variables" inside edge operator and set it inside the edge dictionary (next to the weigth and other parameters), pyrates didnt like it.

So, since the weights of the edges have to be fixed, there is no way to vary them in time, except to design and run the model again, so a bit more manually? Or is there a way with some internal settings?

A similar topic has already been discussed inside the "Continuous data generation" issue, but it was only discussed in the relation to the variables inside the node operator..

Thank you!

[Richert]

Hi Nina,

indeed external inputs cannot be placed on edges at the moment. It is a feature to consider for the future, for sure, but currently it's not supported.
What is supported is to define edge operators via differential equations on edges, if it were possible to express the edge dynamics you are interested in via a differential equation.

Another possibility is that you define a node-specific output variable that is scaled by some extrinsic input and that you could use as the source variable of any edge. This way, you could scale all the outputs (rather than the inputs) of a given node in a time-dependent fashion.

Finally, as a work-around, there is always the possibility to define multiple input variables on a node and define separate extrinsic inputs for them (for example in_total = ext_1*in_1 + ext_2*in_2 + in_3). Then you connect the edges to different input variables and achieve a separation like that.

Hope that helps!

Best, Richard

[NinaOmejc]

Hi,

thanks for the last comments! I've been trying to figure out what exactly you meant by your proposals and somehow came up with the version attached below, which works for me. I post it here in case some one is interested or if you have any comments on it. I guess the downside is that for e.g. 20 couplings, I would have 20 new variables + 20 new inputs, but I guess that is inevitable to some degree.

The approach is also great, because now I can use delays with any issue (as compared to separate runs of simulation).

Thanks again for all the help!

test_kuramoto.zip

[Richert]

Hi Nina,

well if you really want to control 20 different incoming connections via 20 distinct extrinsic signals, that is true. However, if you, for example, want to apply one extrinsic input pattern to one group of incoming connections, and a different input to the second group, you would still be fine with 2 new variables + inputs.

Of course, always happy to help!