Took longer than expected to get the time to make the pull request, but as per request #181 (comment)

This variant will allow you to pass in a variable such as a python object in the callback. Example use case here

def train_blocking(self):
	...
	#python object
	summary = {}
	summary['elapsed'] = stop-start

	#call back with results on game thread
	ue.run_on_gt(self.training_complete, summary)

...
#actual function getting called on game thread with data
def training_complete(self, summary):
	#not related, but forward summary python object to blueprint layer as json string
	self.uobject.OnTrainingCompleteFunction(json.dumps(summary))

https://github.com/getnamo/tensorflow-ue4/blob/master/Content/Scripts/TensorFlowComponent.py#L164

Note that this particular call uses an alias defined here: https://github.com/getnamo/UnrealEnginePython/blob/b652cbfd759da6045a2f89c4d4080526b33d3c47/Source/UnrealEnginePython/Private/UEPyModule.cpp#L501, alias not required for functionality and it's not included in the pull request. Instead you would write:

ue.create_and_dispatch_when_ready(self.training_complete, summary)

For the other route (game thread->run something on background thread) it's probably better to use python's Thread model (see: https://github.com/getnamo/UnrealEnginePython/blob/master/Content/Scripts/upythread.py for an example)