Running mlflow projects in the docker environment does not honour .dockerignore.
This has been an outstanding issue on mlflow for a long time and does not appear to be progressing. We should look at alternatives for running projects. Options:
plain python execution with venv
advantage - simple
disadvantage - uncontrolled
docker run script as entrypoint
advantage - controlled, most similar to the current method
disadvantage - requires some (minor?) MLOps rework, do we lose any of the mlflow project features?
I'm leaning towards option 2 as this offers the best compatibility with existing work. Please feel free to suggest any alternatives or opinions below.
Running mlflow projects in the docker environment does not honour .dockerignore.
This has been an outstanding issue on mlflow for a long time and does not appear to be progressing. We should look at alternatives for running projects. Options:
I'm leaning towards option 2 as this offers the best compatibility with existing work. Please feel free to suggest any alternatives or opinions below.
related to #132