Feature: Whole volume validation

[tomvars]

It would be very helpful to do full volume validation during training. This would involve running one of the Aggregators instead of the Samplers mid-training and using the new niftynet.engine.signals to trigger the evaluation. The result would serve as a template for future signals to run during training.

[wyli]

on a side note there's a very brief overview of niftynet.engine.signal at https://niftynet.readthedocs.io/en/dev/extending_event_handler.html

[ssbpp]

I would like to do full volume validation during training. Is that now possible in some way? It would be really helpful!

[mmodat]

Hi @tomvars Any progress on that? @irme is keen to work on it during the hackathon. @irme and @ericspod - could you have a chat about it and plan what are the required changes?

[ericspod]

@Irme and I have discussed. She has a solution which isn't compatible with just making a PR. We will be changing the code a lot during the hackathon which will make that harder. We feel that figuring out what the solution should afterwards is the way forward.