MLflow Server with Docker

The default artifact storage is configured to use be the folder mlruns.

Setup

For the users' database, please create it following the instructions here or here. Then put the generated .htpasswd file into the config/ folder.

In a nutshell

1. Install apache2-utils

   sudo apt-get install apache2-utils

2. Create at least one user

   sudo htpasswd ./.htpasswd user1

Run the service

1. Build

   docker-compose build

2. Run. Use -d for detached mode (the terminal is not blocked)

   docker-compose up [-d]

Track experiments using mlflow client

Setup

1. Install mlflow

   pip install mlflow

2. Set environment variables:

   Manually:

   # Login information
   export MLFLOW_TRACKING_USERNAME=fill  # Username
   export MLFLOW_TRACKING_PASSWORD=fill  # Password
   # Server information
   export MLFLOW_TRACKING_URI=fill  # Password

   Or in Python. You can copy the previous code snippet into a .env file without the export command and use the dotenv package to load it (pip install python-dotenv).

   • Install required library

     pip install python-dotenv

   • .env file

     # Login information
     MLFLOW_TRACKING_USERNAME=fill  # Username
     MLFLOW_TRACKING_PASSWORD=fill  # Password
     # Server information
     MLFLOW_TRACKING_URI=fill  # Password

   • Python code to load the .env file

     from dotenv import load_dotenv
     load_dotenv(dotenv_path='./.env')

3. Set the tracking URL

   import mlflow
   import os
   mlflow.set_tracking_uri(os.environ['MLFLOW_TRACKING_URI'])

References

• How to run authentication on a mlFlow server?
• Restricting Access with HTTP Basic Authentication