search

aleximmer / Laplace

Laplace approximations for Deep Learning.
https://aleximmer.github.io/Laplace
MIT License
432 stars 61 forks source link

Use backend-native Jacobians if available #187

Closed wiseodd closed 3 weeks ago

wiseodd commented 1 month ago

This is just a hotfix for #186. I will need to talk to Felix for the full fix.

wiseodd commented 3 weeks ago

TODO: Compute Jacobians with the "best" backend during prediction, not necessarily the one used for computing the Hessian in fit. I.e. we should think of the Jacobian (in the GLM predictive) as separate, backend-wise, from the Hessian.

Edit: Should be a separate PR!