Populating Pipeline/Workspace field in kubeflow artifacts

[abhaychand147]

In earlier package for saving metadata we were able to group artifacts based on a workspace, is it now possible with MLMD?

[BrianSong]

Hi,

I am not very clear about the context here as you mentioned "earlier package", could you be more specific on that?

Ideally, if the pipeline/workspace is a field of a artifact, we will be able to group it using the MLMD Declarative Nodes Filtering.

[abhaychand147]

Hi, thanks for replying,

github.com/kubeflow/metadata/blob/master/sdk/python/sample/demo.ipynb

Here in cell3, we can define a workspace and for every pipeline run, all the artifacts will be saved in this workspace.
The same can be seen in kubeflow UI under artifacts section.
Do we have the same grouping option available in mlmd?

[BrianSong]

Hi @abhaychand147, sorry for the late reply(I must have miss it). Unfortunately, the Workspace is not a concept in MLMD, it should be an upstream concept that designed by KFP.

Back to your question, there are two options: 1. use each database as a workspace. 2. store the workspace value in one of the artifact's property and use MLMD Declarative Nodes Filtering to group them.