I spent some time playing with Kubernetes & @kubeflow pipelines, and they have one feature which is just great:
You can define the pipeline with real code!
But there's more.
You can compile the pipeline into YAML and upload it or...
You can use the API to submit it.
Recap:
1. write code to define the steps
2. use code to submit the steps as a pipeline
If you're using a Jupyter Notebook, the experience is seamless.
1. decorate your functions
2. call the pipeline API
3. enjoy your model trained at scale
Since the pipeline is just code, you can express all your logic in Python.
No more clever YAML!
Why isn't writing code over YAML more popular?!
What do you think?
Also...
If you like CI/CD pipelines and machine learning, you should tune in for tomorrow's webinar on Kubeflow with @SoulmanIqbal
This thread is based on his work.
Register here: event.on24.com/wcc/r/2451691/…