Mlflow server with postgres as backend store and minio as artifact store

Inspired by this repo.

Deploy

• go to server where you want to deploy mlflow
• clone repository and move into it:

git clone git@github.com:antklen/mlflow-server.git
cd mlflow-server

• build docker image (only for the first time): sh build_mlflow.sh
• define environment variables with ports (could be any free ports, values here are just for example):

export MLFLOW_PORT=5000
export MINIO_API_PORT=9010
export MINIO_CONSOLE_PORT=9011

• run docker-compose: docker-compose up
• for the first time go to minio console http://{host}:{MINIO_CONSOLE_PORT} and create bucket with name mlflow

Usage

For logging experiments in python:

import os
import mlflow

os.environ['MLFLOW_S3_ENDPOINT_URL'] = 'http://{host}:{MINIO_API_PORT}'
os.environ['AWS_ACCESS_KEY_ID'] = 'mlflow'  # defined in docker-compose.yml
os.environ['AWS_SECRET_ACCESS_KEY'] = 'mlflow_pass'  # defined in docker-compose.yml

mlflow.set_tracking_uri('http://{host}:{MLFLOW_PORT}')

Go to http://{host}:{MLFLOW_PORT} to see mlflow UI.