yijingz/breakeven_complexity
This repository contains the dataset for "Breakeven complexity: A new perspective on neural partial differential equation solvers". Dataset detail: Navier-Stokes: Simulated via Exponax. Contains field "u". 20,000 training trajectories and 1,000 test trajectories. Shape (N, res, res, T). Kuramoto-Sivashinsky: Simulated via Exponax. Contains field "u". 20,000 training trajectories and 1,000 test trajectories. Shape (N, res, res, T). Gray-Scott: Simulated via Exponax. Contains field "u" and "v".… See the full description on the dataset page: https://huggingface.co/datasets/yijingz/breakeven_complexity.
This repository contains the dataset for "Breakeven complexity: A new perspective on neural partial differential equation solvers".
Dataset detail:
- Navier-Stokes: Simulated via Exponax. Contains field "u". 20,000 training trajectories and 1,000 test trajectories. Shape (N, res, res, T).
- Kuramoto-Sivashinsky: Simulated via Exponax. Contains field "u". 20,000 training trajectories and 1,000 test trajectories. Shape (N, res, res, T).
- Gray-Scott: Simulated via Exponax. Contains field "u" and "v". 20,000 training trajectories seperated equally into 10 files, and 1,000 test trajectories. Shape (N, res, res, T).
- BreakFlow: Simulated via PyFR. Contains field "velocityx", ""velocityy", "pressure", "vorticity", and "mask". 2,000 training trajectories and 50 test trajectories for each Reynolds bin. The combined folder is a shuffled collection of the data. The OOD(out of distribution) folder contrains train set of a shuffled collection of bin1 and bin2, and bin3 as test set. Shape (N, T, res, res).
Here are time matching statistics for the datasets:
