MLFlow

If you want to boot up mlflow project with one-liner - this repo is for you.

The only requirement is docker installed on your system

AWS S3 based on this article

1. Install AWS cli Yes, i know that you dont have an Amazon Web Services Subscription - dont worry! It wont be needed!
2. Configure .env file for your choice
3. Configure AWS CLI - enter the same credentials from the .env file

aws configure

AWS Access Key ID [****************123]: AKIAIOSFODNN7EXAMPLE
AWS Secret Access Key [****************123]: wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
Default region name [us-west-2]: us-east-1
Default output format [json]:

4. Create mlflow bucket

aws --endpoint-url=http://localhost:9000 s3 mb s3://mlflow

5. Open up http://localhost:5000/#/ for MlFlow, and http://localhost:9000/minio/mlflow/ for S3 bucket (you artifacts) with credentials from .env file

6. Configure S3 Keys.

For running mlflow files you AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables present on the client-side.

export AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
export AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY

You can load them from the .env file like so

source .env

or add them to the .bashrc file and then run

source ~/.bashrc

7. Test the pipeline with below command with conda. If you dont have conda installed run with --no-conda

MLFLOW_S3_ENDPOINT_URL=http://localhost:9000 MLFLOW_TRACKING_URI=http://localhost:5000 mlflow run git@github.com:databricks/mlflow-example.git -P alpha=0.5

Optionally you can run

MLFLOW_S3_ENDPOINT_URL=http://localhost:9000 MLFLOW_TRACKING_URI=http://localhost:5000 python ./quickstart/mlflow_tracking.py

8. To make the setting permament move the MLFLOW_S3_ENDPOINT_URL and MLFLOW_TRACKING_URI into your .bashrc

export MLFLOW_S3_ENDPOINT_URL=http://localhost:9000
export MLFLOW_TRACKING_URI=http://localhost:5000