At the moment every project needs a run_project.py file and the project is run using python run_project.py ARGS this run_project.py allows flexibility between projects but also introduces some confusion about when to directly run an entrypoint (e,.g train.py) and when to run the project.
I think it will be simpioler to change mlops into a command line tool e.g. csc-mlops run train.py or similar.
it will be important for whatever solution we settle on to keep the flexibility of the current system. including docker args. Kedro looks like a sort of similar project - see if there is anything we can learn from there. https://kedro.readthedocs.io/en/stable/tutorial/create_pipelines.html
At the moment every project needs a run_project.py file and the project is run using
python run_project.py ARGSthis run_project.py allows flexibility between projects but also introduces some confusion about when to directly run an entrypoint (e,.g train.py) and when to run the project.I think it will be simpioler to change mlops into a command line tool e.g.
csc-mlops run train.pyor similar.it will be important for whatever solution we settle on to keep the flexibility of the current system. including docker args. Kedro looks like a sort of similar project - see if there is anything we can learn from there. https://kedro.readthedocs.io/en/stable/tutorial/create_pipelines.html