Nathan Lambert

Make plotting robust

1

The best way to do plotting is to point it to a logged data directory. WIth new sweeping, we need to part plotling code to load directories and see what...

Integrate SAC implementation

1

Right now the code does not work with our dynamics models, we need an environment based model. Also look at the envs in utils to baseline some things.

Consider training on only recent data, or re-weighting in a clever way to improve performance of the MPC.

Integrate work in the Austin Branch, or here https://github.com/austinnguyen517/dynamicsPID more cleanly. - [ ] PID Code setup and confirmation - [ ] BO code in general - [ ] Variable...

Consider the Dagger algorithm http://proceedings.mlr.press/v15/ross11a/ross11a.pdf

The readme is really old

multiple prediction options in ddpm, ddim

13

starting to work on discussion in #778. Please contribute and leave feedback. This is mostly a placeholder for my work right now as I figure out how to do it....

Unused parameters in scheduler `init`'s

12

**What API design would you like to have changed or added to the library? Why?** I see some of the [schedulers](https://github.com/huggingface/diffusers/blob/7bd50cabafc60bf45ebbe1957b125d3f4c758ba8/src/diffusers/schedulers/scheduling_lms_discrete.py#L34) had these parameters added to init ``` trained_betas: Optional[np.ndarray]...

Annealed Langevin Sampler

8

Closes #84 Maybe @MinkaiXu can give a review on this?

Clean up 1d unet blocks

9

Some changes motivated by #105.