This example is based on canonical/kubeflow-example/e2e-wine-kfp-mlflow with some modifications to simplify the codebase. This example is explained in more details in this blog post.
You should first navigate to the kubeflow dashboard at $IPADDR.nip.io and
create a jupyter notebook server under the "Notebook" tab in the sidebar. Then,
you can choose an appropriate cpus and memory for your server; the
default value is good enough for this demo. Also, check
Allow access to MinioAllow access to MLFlow
so that the necessary environment variables will be imported to the notebook server.
You can either choose to run the demo with the python script by first installing
the required package pip install -r requirement.txt, and then run python3 pipeline.py to build, train, and deploy the model. Or you can follow the
notebook e2e-kfp-mlflow-seldon-pipeline.ipynb that basically do the same thing
as the python script.
After running the script or notebook, you can view the result in the "Runs" tab and view the model in mlflow dashboard. For more information about demo, you can visit the blog post.
Lastly, you can try to make inference of new data using the script sample-prediction.sh.
microk8s kubectl get all -n admin
microk8s kubectl get sdep -n admin