Datalake - Dataset creation

From the Datalake, you will at some point need to create a DatasetVersion containing Data that have been previously uploaded to Datalake.

1. Select the Data

To create a Dataset on Picsellia, you first need to select from the Datalake the Data that will compose your Dataset. To do so, you can rely on the ordering, filtering, and selecting features on the Datalake detailed previously.

This is where every exploration tool covered so far comes together: the Query Language (searching on Tag, fixed Metadata, or any other property), Charts, and Embeddings exploration are all different ways to navigate and pinpoint the relevant Data inside a large Datalake, so you can select exactly what belongs in your ideal Dataset.

Once Data are selected in the Datalake, a Dataset dropdown button appears next to Process, Tag, and More:

2. Dataset creation

Picsellia proposes a Dataset Versioning system, which is detailed here, but basically, it means that each Dataset can have several versions (called DatasetVersion).

The Dataset dropdown proposes three options:

• "Create New Dataset": Creates a new Dataset and its first DatasetVersion (named initial by default) with the selected Data. A modal opens where you provide the Dataset's name and an optional description.

• "Create New Dataset Version": Creates a new DatasetVersion of an existing Dataset with the selected Data. A modal opens to pick the Dataset from a grid of your existing ones, then name the new DatasetVersion and optionally describe it.

• "Add To Existing Dataset Version": Appends the selected Data to an existing DatasetVersion. A modal opens to select the Dataset and DatasetVersion the Data will be added to.

  

You can now jump to the Dataset tab of Picsellia to visualize the freshly created DatasetVersion.