A framework to layer a standard interface over datasets commonly used in machine learning.
- Store datasets on disk in their canonical format, for interoperability
- Transparently download datasets on first use
- Handle common dataset tasks:
- Splitting datasets
- Data preprocessing
- Minibatching