thuerey-group/apebench-scraped

---
license: mit
---

# APEBench Scraped

A representative subset of datasets created using the [APEBench benchmark suite](https://github.com/tum-pbs/apebench) using version `0.1.0`.

⚠️ Note that [APEBench](https://github.com/tum-pbs/apebench) is designed to procedurally generate all its training and test data. This allows for advanced features like benchmarking approaches with differentiable physics. Hence, there is no need to download this dataset as it can be easily re-generated using APEBench which can be installed via `pip install apebench`. See also [here](https://tum-pbs.github.io/apebench/examples/scrape_datasets/) for how to scrape datasets.

## Download

Download without large files

```bash
GIT_LFS_SKIP_SMUDGE=1 git clone [e-mail]:datasets/thuerey-group/apebench-scraped
```

Afterwards, you can inspect the repository and download the files you need. For
example, for `1d_diff_adv`:

```bash
git lfs install
git lfs pull -I "data/1d_diff_adv*"
```

Alternatively, you can download the entire repository with large files (~30GB):

```bash
git lfs install
git clone [e-mail]:datasets/thuerey-group/apebench-scraped
```

## Reproduction

Obtained via:

```bash
conda create -n apebench python=3.12 -y
conda activate apebench
pip install -U "jax[cuda12]"
pip install apebench==0.1.0
```

Alternatively, you can use the provided `environment.yml` file:

```bash
conda env create -f environment.yml
conda activate apebench
```

And then executed the following script (also found under `reproduce.py`):

```python
import apebench
from tqdm import tqdm
import os

DATA_PATH = "data"

os.makedirs(DATA_PATH, exist_ok=True)

for config in tqdm(apebench.scraper.CURATION_APEBENCH_V1):
apebench.scraper.scrape_data_and_metadata(DATA_PATH, **config)
```

⚠️ Small️️er variations of the generated data can occur due to different JAX
versions, backends (CPU, GPU, TPU), drivers, etc. This might be especially
pronounced for the chaotic problems (like KS or Kolmogorov flow).

- nvidia driver version: 535.183.01
- cuda version: 12.2
- GPU: RTX 3060