cyyeh / ml-platform

ML Platform

idea credit: https://github.com/erdiolmezogullari/ml-fmnist-mlflow-tensorboard

In this case, we used docker container technologies to create the ML platform from scratch. It consists of four different docker containers (MLflow, JupyterLab, PostgreSQL, TensorBoard) that are already built in docker-compose.yml. Now we installed TensorFlow2.3.1 in JupyterLab.

Prerequisites

• Install Docker
• Install docker-compose
• If you would like to use GPU, please install nvidia-docker2, and set nvidia as the default runtime
  • add "default-runtime": "nvidia" to /etc/docker/daemon.json like this:

  {
    "default-runtime": "nvidia",
    "runtimes": {
        "nvidia": {
            "path": "/usr/bin/nvidia-container-runtime",
            "runtimeArgs": []
        }
    }
  }
  

  • execute sudo systemctl daemon-reload and sudo systemctl restart docker

Usage

See this example project for demonstration!

1. Add the training data under the data directory.
2. Add training scripts, notebooks under the app directory.
3. Experiment artifacts would be stored under the artifacts directory.
4. Database logs and Tensorboard logs would be stored under the logs directory.

• To start the platform(it will pull the latest images of the ml-platform from Docker Hub):

    make up
  

• To stop the platform:

    make stop
  

• To stop the platform and remove containers:

    make down
  

• To visit JupyterLab, type the following address on your favorite browser:

    http://localhost:8888/
  

• To visit TensorBoard, type the following address on your favorite browser:

    http://localhost:6006/
  

• To visit Mlflow, type the following address on your favorite browser:

    http://localhost:5000/
  

• To build the platform(only needed if you want to customize it), update the image tag for each service you would like to update in docker-compose.yml file and then:

    make up
  

Notes on docker image sizes(statistics based on 0.1.0 version)

• Before minimizing docker image sizes
  • ml-platform_jupyterlab: 6.26GB
  • ml-platform_mlflow: 1.15GB
  • ml-platform_tensorboard: 1.47GB
  • ml-platform_postgres: 0.16GB
• After minimizing docker image sizes
  • ml-platform_jupyterlab: 3.7GB(41% reduced)
  • ml-platform_mlflow: 1.09GB(5% reduced)
  • ml-platform_tensorboard: 0.21GB(86% reduced)
  • ml-platform_postgres: 0.16GB(0% reduced)