Datasets

Datasets#

datasets.tutorial_data([path])

Download, verify and cache the openDVP tutorial dataset.

Where the data lives#

tutorial_data() downloads a single archive from Zenodo (10.5281/zenodo.15830141), checks it against a known MD5, extracts it, and returns a dict of paths. The download happens once and is then cached per-user, so the tutorials are re-runnable offline.

The cache location defaults to your platform’s user cache directory — ~/Library/Caches/opendvp on macOS, ~/.cache/opendvp on Linux. Override it either per-call or for the whole environment:

import opendvp as dvp

paths = dvp.datasets.tutorial_data(path="/scratch/opendvp-data")
export OPENDVP_DATA_DIR=/scratch/opendvp-data

The archive is 133 MB compressed and roughly 1 GB extracted, almost all of it the multiplexed image.