nitroml
nitroml copied to clipboard
[Datasets] Create Dataset download component.
We need a way for TFDS to lazily download and prepare datasets, so that it only runs at pipeline execution time instead of during DAG construction time.
Do you think we still need this? We can change the TFDS data_dir (e.g. https://github.com/googleinterns/nitroml/blob/nitroml_cloud/examples/titanic_benchmark.py#L50)