jupyter notebook Link to heading

Notebook components are installed by default in example manifests.

Just simply follow the UI, you will be able to create a jupyter lab server.

I use this repo to demonstrate the usage of kubeflow.

model registry Link to heading

Model registry is not installed by default. However this is an very important component for me in an ideal machine learning platform.

To install model registry is very simple - just follow this documentation to install.

One thing worth mentioning is: if you choose to install model registry to kubeflow namespace (what PROFILE_NAME regulates), you won’t see anything useful in UI (see this issue). Instead, assign PROFILE_NAME to the namespace you use, in my case kubeflow-user-example-com, you will be able to validate model registry thru UI.

minio Link to heading

Minio is a surprise to me. I was always looking for an open sourced S3 like tools, and there it is - minio. In example manifests, it will install minio by default. But if you want to see minio UI, you will still need to port forward by using something similar to kubectl port-forward svc/minio-service -n kubeflow 9000:9000. After portforwarding, go to localhost:9000 to check. kubeflow project set minio username as minio, passowrd as minio123.

all pieces together Link to heading

1. After jupyter lab server started, you can download the repo into the workspace, pip install then run all cells.
2. In the ipynb, I wrote a not-so-strict idempotent register function to register the model; once registering is successful, it will also upload model artifact into blob storage (minio in this tutorial); if registering is not successful, it will not upload the model into storage.
3. model registering consequences can be visualized in model registry UI.
4. model artifact can be found in storage UI.