Tags

As an Organization grows, it accumulates dozens of Dataset, Project, Model, and Deployment, and their names alone no longer tell you which ones are the reference benchmarks, which ones belong to a given customer, or which ones are still experimental. Tags give your team a shared vocabulary to organize these objects and retrieve them in a few clicks, from any list of the platform.

This page covers the tags of the platform objects and how they are managed for the whole Organization. The tags attached to the content of a Datalake, a DatasetVersion, or a Deployment have their own dedicated pages, linked below.

1. Two families of tags

A tag is a name, with an optional description, that you attach to one or more objects to structure them and filter on them. It is not a Label and has no effect on training or on predictions. Picsellia distinguishes two families of tags:

• Tags on the content of a container: the DataTag of a Datalake, the AssetTag of a DatasetVersion, and the PredictedAssetTag of a Deployment. They are created and managed from the Settings of their container, and each container has its own independent set. They are detailed in Datalake tagging system, DatasetVersion tagging system, and Deployment tagging system.
• Tags on the platform objects: the tags of Dataset, DatasetVersion, Project, Experiment, Model, ModelVersion, Deployment, and Processing. They are shared across the whole Organization and managed from its Settings, as detailed below.

Each type of object has its own set of tags: a tag created for Dataset can only be attached to a Dataset, not to a Model or a Project. This keeps each list filterable on tags that make sense for it.

2. Manage tags from the Organization settings

The Tags page of the Organization settings lists, in one place, every tag of the platform objects. As for the rest of the Organization settings, only the Admin of the Organization can access it (see Configure an Organization).

The page holds one tab per type of object: Datasets, Dataset Versions, Projects, Experiments, Models, Model Versions, Deployments, and Processings. On each tab, the counter above the list gives the number of tags of this type, and the search field filters them by name.

To create a tag, click on the + Add new … tag button of the tab (for instance + Add new dataset tag), type its name and, optionally, a description, then confirm with the check button. Defining the tags here, before anyone starts tagging, is how you keep the vocabulary consistent across the team instead of ending up with benchmark, Benchmark, and bench side by side.

To delete tags, tick them in the list and click on the trash icon next to the search field. From this page, you can also edit the name and description of an existing tag.

3. Attach and detach tags from a list

Tags are attached and detached from the list of the objects themselves: the lists of Dataset, Project, Model, Deployment, and Processing, and the lists of DatasetVersion, Experiment, and ModelVersion displayed inside their parent object.

Tick one or more objects in the list: the number of selected objects is displayed above the list, next to a Tag button. Click on it and choose Attach or Detach.

The Attach modal lists the existing tags of this type of object, with a search field to find them by name. Tick the ones you want and click on Attach. If the tag you need does not exist yet, the Create new tag row at the bottom of the modal creates it on the fly, with its name and an optional description, without going through the Organization settings.

The Detach modal works the same way and only lists the tags actually attached to your selection, so you can see at a glance what will be removed.

4. Filter a list by tag

Once tags are attached, they are displayed with each object in the list, and the Select Tags button above the list filters it on them. Tick the tags you want in the popover and choose how they are combined before clicking on Apply:

• AND: displays the objects that carry all the selected tags.
• OR: displays the objects that carry at least one of the selected tags.
• NOT: displays the objects that carry none of the selected tags.

Select all ticks every tag of the list at once, and Reset clears the filter. It is the same filter as on the Data of a Datalake, detailed here.