Made pybnn and torch optional

automl / RoBO

RoBO: a Robust Bayesian Optimization framework

BSD 3-Clause "New" or "Revised" License

483 stars 133 forks source link

Made pybnn and torch optional #103

Closed KOLANICH closed 5 years ago

aaronkl commented 5 years ago

Would you mind explaining why pybnn should be optional?

KOLANICH commented 5 years ago

Because it is used only in wrapper_bohamiann.py. The rest of the lib doesn't need that. And these are dependencies, some prefer not to have on own machines, because it is too costly to build and/or keep them.

KOLANICH commented 5 years ago

@aaronkl

aaronkl commented 5 years ago

one of the key components of RoBO compared to other BO packages is that it support Bayesian neural networks (e.g Bohamiann and DNGO) which are implemented in the pybnn package. So I don't see why we should make that optional. For instance, it is also needed in the central fmin and mtbo interface

KOLANICH commented 5 years ago

No, it is not needed. Some parts are fully useful without pytorch. I had to patch the lib and I don't think it is a right solution.