domino-datasets

Dataset versioning captures previous data states using snapshots to ensure reproducible machine learning experiments. Snapshots allow you to rollback changes during experimentation, enabling you to recreate exact training, validation, and production data states for reliable testing.

To create a versioned dataset, initialize a dataset and enable snapshot versioning. The data is then mounted under /domino/datasets/ for fast access, providing scalable and persistent storage that streamlines data management across your project workflows.

Yes, cross-project sharing allows you to mount datasets across different teams and projects. This streamlines collaboration by enabling multiple projects to access the same versioned, high-performance storage, reducing data silos and duplicated data pipelines.

Yes, Domino datasets support scalable mounting specifically for ML workflows. They mount directly under /domino/datasets/ to provide fast, persistent storage access, satisfying the requirements of scalable data engineering and machine-learning pipelines.

The best way to manage high-performance datasets is using versioned storage with snapshots. This approach ensures reproducibility by capturing specific data states, allowing you to rollback changes and maintain separate datasets for training, validation, and production.

You should avoid shared dataset approaches if your workflow does not require cross-project collaboration or state reproducibility. Without the need for snapshot rollbacks or persistent mounting under /domino/datasets/, standard storage may better suit isolated, single-use data tasks.