[ENH] Add cuda support for all deep learning models in library (PyTorch, Jax) #22
Description
Problem description
Hi! There are some optimizers and functions in library based on usage of deep learning models. I think we should add cuda support for all of this models.
Proposed solution
Add cuda computations support for all surrogates and functions, which use neural networks.
List of surrogates:
- DeepGP,
waggon.surrogates.DGP - Gan,
waggon.surrogates.WGAN_GP - BNN,
waggon.surrogates.BNN - DE,
waggon.surrogates.DE
List of functions:
- MLP hyperparams,
waggon.functions.MLSolver, - NN hyperparams,
waggon.functions.NNhyperparams
To discuss
Execution device manually setup
Should we add on option to manually configure the execution device?
I think in future we should add device parameters to the surrogates, which takes following values:
'cpu'-> use only cpu'auto'-> use gpu if available
Execution device setup for functions
How we should use GPU for functions?