Batch simulation: batching over deeper Net params?

[darcywaller]

Hi team - I was checking out the new BatchSimulate function in 0.4 and it seems it is set up primarily for testing different parameters of the external drives. Is there any functionality or can you suggest the best way to test multiple parameters for the local network parameters (e.g., let's say, L5 basket -> L5 Pyr soma GABAb weights)?

A challenge I am facing is that I can set up a new network configuration in a notebook and test it for such a parameter change but as referenced in #774 (comment) , such an externalized network instance won't simulate in parallel because of some MPI security settings, and have to manually change the hnn-core code in my own fork to test out new network configurations if I want to run multiple trials or optimize in parallel. Would be great to 1) not have to change the hnn-core library code each time I want to sub in a parameter to test or 2) be able to batch this at once over a range of values in parallel.

[ntolley]

This isn't an answer to your question, but in the short term would it make sense to parallelize with joblib? If you're using batch simulate then at some level each simulation in the batch independent from the other.

[ntolley]

Also you can configure batch simulate to update internal network values in a custom set_params function:

def set_params(param_values, net):
    net.cell_types['L5_pyramidal'].your_attribute_here = param_values['custom_value']

param_dict = {'custom_values': np.linspace(0, 5, 10)}

[darcywaller]

Thanks Nick, that ended up working just fine - assumedly because the parameter adjustments on the network are happening within the scope of the hnn-core scripts and not out in an external notebook.