Passing objects through the stack

[NicolasGensollen]

When writing a stack for the metric computation, I faced the following situation:

layer network_split:

create a network_analyzer object with the model received from previous layer:

net_analyzer = network_analyzer(model) #This is kinda slow...

and do some modifications:

net_analyzer.do_something() #Modifs that split the network into feeders

and finally returns:

return network_analyzer.model

layer compute_metrics

Receives the model from the network_split layer, but has to re-create an instance of network_analyzer to compute the metrics:

net_analyzer = network_analyzer(model) #This is (again) kinda slow...
net_analyzer.compute_metrics()
net_analyzer.export()

This means that we end up doing something slow twice when running the stack.

A quick and dirty fix I made was to allow the metric computation to happen in the network_split layer (see bd5e8f5), but it'd be nice to have a cleaner solution. Especially since we need to have a layer dedicated to the metric computation itself. It looks like being able to pass the network_analyzer from one layer to another through the stack (like what is done with the model) would solve this.

I an unsure about the feasibility and complexity of such a change though. And even if possible, this might be something that we don't want to allow.

Thoughts??

[elainethale]

I think it is generally reasonable to have the stack serve as a data repository for the run ... How high priority is it for you all?

[NicolasGensollen]

@elainethale I wouldn't say this is top priority for what we are currently doing since there is a way around that seems to work fine. But it might be nice to have this at some point.