Fink ML containers

Scripts to automatically create containers based on MLflow artifacts of an experiment. The model needs to be registered. Controller usage:

./fink_container -h

Fink ML containers: get artifacts from MLflow, build Docker image, run Docker container

Usage: fink_container [OPTIONS]

Options:
  -h                  Display this help
  -e                  Experiment name, as written in MLflow
  -topic              Kafka topic from Fink with alerts to process. Only used if --run_image is specified.
  --list_experiments  If specified, list available experiment names.
  --artifacts         If specified, get artifacts (experiments and models) from MLflow
  --build_image       If specified, generate Dockerfile and build image based on artifacts
  --run_container     If specified, run a container that process alert data.

Examples
  # List experiment names
  fink_container --list_experiments

  # Get artifacts
  fink_container -e toto --artifacts

  # Build Docker image
  fink_container -e toto --build_image

  # Run container
  fink_container -e toto -topic fink_topic --run_container

Todo:

• finish to write science module example with full consumer/producer
• automate the deployment of the container. mlflow serve?
• better use of the mlflow client?