-
CUJ.The CUJ is to track model development in Notebook and Kubeflow Pipeline. (deck)
-
API and metadata specifications. The goal is to define API and a set of built-in metadata specifications that can support the CUJ above; The API should be able to handle customized metadata specifications.(issues #4 #5)
-
Implementations. We plan to leverage MLMD for an implementation of the API, because it is already used by Kubeflow Pipeline. Different implementations are welcome and should be easily discoverable.(issues #8 #9)
-
Python library. We plan to create a Python library for logging metadata in Notebook and Kubeflow Pipeline.(issue #6)
-
User interface. We plan to have a page in Kubeflow dashboard to show models and workflows in Notebook and Kubeflow Pipeline. (issue #7)