Review learn related framework capabilities for our new architecture
In order to move on to a more detailed polaris architecture and make Polaris evolution easier we are planning to reframe how we create learning components in polaris. Several frameworks need to be review for that purpose:
- Scikit-learn pipelines
- Luigi tasks graphs https://github.com/spotify/luigi
- MLFlow https://mlflow.org/
- whichever...
And how they can work together Lugi is sort of a task graph while MLFlow is a tracker of models to augment reusability of what works the best for instance.
This is issue is to research that, output is decision of what we would implement.