Is there an easy extension for SGMCMC such as SGLD and SGHMC?

AdamCobb / hamiltorch

PyTorch-based library for Riemannian Manifold Hamiltonian Monte Carlo (RMHMC) and inference in Bayesian neural networks

BSD 2-Clause "Simplified" License

426 stars 63 forks source link

Is there an easy extension for SGMCMC such as SGLD and SGHMC? #25

Closed timxzz closed 1 year ago

timxzz commented 1 year ago

Hi Adam!

I found this repo very useful! I am wondering if we can also have the functionality of SGLD and SGHMC? Or if I can change some params to directly achieve that already?

Thanks a lot!

Best, Tim

AdamCobb commented 1 year ago

Hi Tim,

Sorry for the delayed response and thank you for reaching out!

There is currently not a functionality for implementing SGLD and SGHMC. It might be possible to do so with Hamiltorch but I recommend going to the original papers to implement these. Hamiltorch is generally focused on non-stochastic versions of HMC, but that is not to say that it wouldn't have it in the future.

Best,

Adam

timxzz commented 1 year ago

Hi Adam,

Thanks for the reply and the suggestion!

Best, Tim