My view on how MLOps should be implemented on Microsoft Azure

1. Make changes to train.py
2. Run the model training on local (possibly downsampled) version of the data
3. Update pipeline.py and any unit tests
4. When you are ready to train the model on the training cluster, tag the branch with, e.g., git tag -a -m "Training with new hyperparams" train-vN.N
5. Push the branch with the --follow-tags flag set, so git push --follow-tags origin master