NeuronUnit Optimization Light build.

[russelljjarvis]

@rgerkin

The two backends RAWpy, hhrawf.py only depend on numba jit deap and dask. And all of these packages seem to play nice with pip, and don't necessarily need apt-get level intervention.

The optimizer didn't really depend on BluePyOpt (I have commented out this module), as it only uses two files from BPO (bluepyopt/deapext/ algorithms.py, and optimization.py [now integrated with NeuronUnit]), which I have heavily customized and they do not or should not depend on any existing BPO code (BPO is dependency heavy).

Therefore it might be possible to create a lightweight pip installable version of the NUOptimizer, that does not depend on NEURON, PyNN, BluePyOpt, PyNeuroML or anything, big heavy and C external library based. We also don't want begginers to have to use apt-get.

The later docker layers should definitely include these big heavy external C based programs, with all the support they offer.

Having a highly pythonic pip installable version could mean that it's easier for beginners to pick this up.

[russelljjarvis]

Also to help this along, my latest git commit does away with the dependency on BluePyOpt, and I created an __init__.py for models/backends, called __light_backends__.py. This will facilitate a really light weight pip installable NUOpt env.