Set something up so that optimisation/inference can be stopped and restarted

[MichaelClerx]

The ask-and-tell interface (#132) takes care of part of this, but it'd also be good to store the internal state of our optimisers/inferers to disk, perhaps using pickling ?

[MichaelClerx]

#TODO: Have a quick look at https://pythonhosted.org/joblib/

[MichaelClerx]

Meh

[MichaelClerx]

@martinjrobins @ben18785 please add Freya to this ticket if she decides to take it up :-)

Rough idea of how to implement it:

• write it as some kind of "deep copy" of every sampler/optimiser's state
• add bits to the ask-and-tell interface that allow the current state to be obtained, and a previous state object to be restored
• add a default implementation using copy.deepcopy
• but make sure that copy.deepcopy is not required !
• figure out what to do for wrapper methods i.e. cmaes

[martinjrobins]

@MichaelClerx : points 3 and 4 seem contradictory, what do you mean that copy.deepcopy is not required, its part of the standard library isn't it?

[MichaelClerx]

Sorry, I meant that the interface shouldn't be written such that implementations require deepcopy. I.E. it should be possible to write a custom implementation instead

[MichaelClerx]

Via a different route I've hit on https://joblib.readthedocs.io/en/latest/ again :D