Skip to content

Dataset licensing

TimeNet's own code is MIT licensed. That license does not extend to the datasets TimeNet fetches and converts. Each dataset keeps the license set by whoever published it.

Where a dataset's license lives

Every dataset ships a card (dataset.yaml) with fields that tell you what applies:

  • license: an SPDX identifier, such as MIT, Apache-2.0, or CC-BY-4.0. The dataset-card schema limits this field to a fixed set of values, kept in sync with timenet.types.License. Most are license identifiers you can look up on SPDX. Two are escape hatches:
    • OTHER: a known license outside the SPDX list, such as a Hippocratic License 3.0 build with modules. Pair it with a license_url.
    • NONE: SPDX's value for a source whose publisher states no license at all. It does not mean the data is free of restrictions: its terms of use are unknown, so check with the publisher before you use or redistribute it.
  • license_url: where to read the full license text. Required when license is OTHER.
  • source_url: where the data comes from. Open it to read the upstream terms in full. This field is optional, so a purely synthetic dataset may leave it out.

The dataset catalog lists both fields per dataset.

Before you download

Some sources only grant credentialed access. PhysioNet, for example, asks you to accept a data use agreement and sign in before you pull a record. TimeNet does not remove that step. When a dataset's source_url points at a gated source, follow that source's terms and use your own credentials.

Redistribution

TimeNet does not vendor dataset bytes. It fetches each dataset from its source_url, so you get the data under the license the source grants. If you redistribute a converted dataset, keep its original license with it.