Downloading the dataset

The dataset is published as a dataset.tar.gz asset attached to each tagged release of the FLARE GitHub repository. There are three common ways to download it.

Tip

You can browse the dataset contents in Problems or GitHub without needing to download the dataset.

Python (recommended)

formulation-bench ships a download_dataset helper that fetches the tarball, extracts it in a local cache directory, and returns the extracted path. Subsequent calls reuse the cached copy.

from formulation_bench import download_dataset

root = download_dataset()                   # latest version this package targets
root = download_dataset("dataset-v0.2.0")   # download a specific release tag
root = download_dataset(force=True)         # re-download, overwriting cache

It is often more convenient to use Dataset.load():

from formulation_bench import Dataset

ds = Dataset.load()

Cache location, in order of precedence:

1. cache_dir= argument to download_dataset / Dataset.load.

2. $FORMULATION_BENCH_CACHE environment variable.

3. $XDG_CACHE_HOME/formulation_bench if $XDG_CACHE_HOME is set.

4. ~/.cache/formulation_bench.

curl

If you just want the dataset files on disk, fetch the tarball directly from the release page:

VERSION=dataset-v0.2.0
curl -L -o dataset.tar.gz \
    "https://github.com/henryrobbins/flare/releases/download/${VERSION}/dataset.tar.gz"

mkdir -p formulation-bench && tar -xzf dataset.tar.gz -C formulation-bench

Warning

The archive expands to a top-level dataset/ directory, so running tar -xzf in a working directory that already contains dataset/ will overwrite it. It is recommended to extract it in a fresh directory (like above).

You can now load the dataset with:

from formulation_bench import Dataset

ds = Dataset("formulation-bench/dataset")

GitHub website

1. Open the FLARE releases page.

2. Pick a release (e.g. dataset-v0.2.0).

3. Under Assets, click dataset.tar.gz.

4. Extract it with your archive tool of choice.