- In this simplistic example, combining mlflow and github-actions to get closer to an automated pipeline that won't cause software engineers to smash their faces against their keyboards
Objective:
- When my model improves, check my code into Github to share with my make-believe team.
MLflow gives me:
- the experiment tracker (for me, as a noob, a UI where I can see if the model iz gooder or not)
- a packaging tool so that I reproduce my project on another machine (mlflow run this thing vs. a bunch of linux and python commands)
Github gives me:
- Google docs for code (tracks every change with versions so I can see who did what and rollback if needed)
- Github Actions, which automatically builds and runs my project on each code change. So even though I pinky swear this code works on my machine, GH-Actions will verify by spinning up some compute in the cloud and running my code
- I also have my hyperparameters (one key lever you can pull on to improve the model) in my gh-action config file, to make it easier to see exactly what is changing between experiments
I'm only scratching the surface, but it's nice to see how well these tools work together to help build an automated ML pipeline.
#mlflow #python #machinelearning
