Make minuit instance once for multiple hypotheses

[dachengx]

A Minuit instance is initialized every time we call StatisticalModel.fit, in some cases when there are a lot of hypotheses, the computational efficiency is not so high.

alea/alea/statistical_model.py

Lines 243 to 245 in 80f6b16

	m = Minuit(
	MinuitWrap(cost, s_args=self.parameters.names),
	**defaults)

[dachengx]

Well, if the toydata is generated once per iteration, and multiple hypotheses are required, we should initiate the Minuit once, because it is only a function of toydata set and likelihood definition.

[hammannr]

It's true though if I remember correctly it caches some of the gradients. Do you have an estimate of whether this is a sizeable effect? Only then I would consider reusing the same instance multiple time (we should anyways time all of this at some point (#20 ).
If we want to implement it we could do it by defining self.minuit_object = None in the init, only generate it when it's not None and set it to None if we assign new data or something 😊