Datasets
Datasets is one of the three workspaces, alongside Projects and Administration. It lists every dataset you can reach, without you having to choose a unit or a project first.
A dataset is a file you can work with, held outside any one project. The Concepts page explains why you might use one; in short, datasets can be shared, versioned, and have their metadata indexed.
Datasets and versions
A dataset is the thing that is named and shared. A version is the thing actually displayed: every dataset has one or more versions, and a URL naming a dataset alone resolves to its current version, so what you are looking at is always a specific version.
The index lists each dataset with its file name, its labels, its editors, its versions and the number of projects it is attached to, and each dataset's versions can be expanded beneath it. The list is narrowed by Search datasets, and by filters for owner, editor, label and file type.
Uploading a dataset
Upload dataset above the list opens Upload New Datasets, where you drop the files you want and send them together.
The dialog asks for a Billing unit — the unit that will pay for the datasets' storage. This is the one place a unit is asked for outside project creation, and it is asked for as billing context rather than as a scope: it does not decide what you can see. Every file in one batch is billed to the unit that batch started with, and the unit you last uploaded successfully in is offered again next time.
Where you may not upload, the control states why rather than disappearing.
Working with a dataset version
Selecting a version opens its details, which hold:
- Dataset Actions - creating a new version of the dataset, which succeeds its latest version rather than the version you happen to be looking at.
- Editors - who may change the dataset.
- Labels - the dataset's own labels, which the index can be filtered by.
- Working Version - which version you are looking at, and downloading it.
- Information - the version's own metadata.
- View - opening the version in the Plaintext Viewer or the Browser Viewer.
- Actions - attaching this version to a project, viewing and editing the dataset schema, and deleting this version.
Attaching a version to a project
Attach Dataset to a Project puts a dataset version into a project's files, so everything you run in that project can read it. The dialog asks for:
- Project - the project to attach it to. This lists the projects you can write to.
- File Type - how the file should be understood.
- Path - where in the project the file should appear.
- Immutable - whether the attached file may be changed in the project.
- Compress - whether the file should be compressed.
A project's Files section can detach an attached version again, and can create a new dataset from a project file. Attaching is carried out by the Data Manager on your behalf, so it appears as a task in the project's Results.
Deleting
Delete this Version of the Dataset removes one version. Deleting the version you are looking at takes you to the version that remains, or back to the index when none does. Versions can also be selected in the index and deleted together.