David Pitt comments

Results 57 comments of


                                            David Pitt

Is it possible to predict multiple timesteps?

Multiple timesteps into the future as in, given a(x,t), predict u(x,t+2)? Or as in given a(x,t) predict u(x,t) and u(x,t+1) at the same time?

Is it possible to predict multiple timesteps?

The library doesn't currently have a training module written for the autoregressive use case

Add a transformer neural operator model and an accompanying example for training it on Darcy flow

@zijieli-Jlee thanks for the edits, your PR is in good shape. if you fix the conflicts we can re-run the tests and see if it's ready to go.

Add a transformer neural operator model and an accompanying example for training it on Darcy flow

Thanks for the fixes @zijieli-Jlee ! I fixed a one-line conflict with a model import since the last update to main. This PR is looking great overall, but I think...

Add a transformer neural operator model and an accompanying example for training it on Darcy flow

Looks great! Asking @JeanKossaifi for final approval.

modernized examples

Can you be more specific? The example scripts in `scripts` are up to date.

Added test for uno horizontal_skips_map when None, and make validate_scaling_factor to return None if all scalings are 1 for eff.

Superseded by #400, closing

Added test for uno horizontal_skips_map when None, and make validate_scaling_factor to return None if all scalings are 1 for eff.

Superseded by #400

Bbonev/local neuraloperators

Superseded by #467, thanks @bonevbs !

Using TFNO to train darcy-flow dataset parameter settings is inconsistent

Hi @Yang-Changhui , for standard model hyperparameters it's simplest to refer to the original FNO paper, linked [here](https://arxiv.org/abs/2010.08895).