This README is still work in progress

Machine learning competition platform

• Machine learning platform just like private Kaggle written in Ruby on Rails framework.

Overall Architecture

Demo environment

URL

https://demo-ml-competition-platform.herokuapp.com/

Demo user account

ID	[email protected]
Password	^dvTT0JgIPS4V@RWW$Kw

Sample JSON file for submission

• AUC Competition
• mAP Competition

Developer's manual

• Developer's manual is written in Github Wiki Page (Still WIP)

How to setup development environment

Deploy API server

• Clone codes of API Server to calculate metrics for competition

  git clone https://github.com/AillisInc/ml_competition_platform.git

• Deploy API server according to the following steps

  • https://github.com/AillisInc/ml_competition_api_server/blob/master/README.md

Setup environment variables.

cp .env.sample .env
vim .env

Copy the following text to .env file.

METRICS_API_ENDPOINT=http://host.docker.internal:8000/
METRICS_API_AUTH_KEY=secret_key  # In production environment, a strong key is recommended.

About METRICS_API_AUTH_KEY

• METRICS_API_AUTH_KEY is used for authorization with API server to calculate metrics
• Therefore, in production environment, it is recommended to use METRICS_API_AUTH_KEY value that cannot be easily guessed.
• METRICS_API_AUTH_KEY value should be matched to API_KEY_TOKEN value in API server

Build docker container with docker-compose

$ docker-compose run web bundle install
$ docker-compose run web rails db:setup
$ docker-compose up

Create User Account

• create admin account with the following command in rails console.

User.create(email: '[email protected]', password: 'password', name: "admin_user", role: "admin")

• in case of creating normal user account, execute the following command.

User.create(email: '[email protected]', password: 'password', name: "test_user", role: "member")

Server URL for development environment

http://localhost:3000

Web API to calculate metrics

• To calculate metrics to determine ranking of each competitor, Web API written in Flask should be also deployed.
• The github code of this Web API server is here

License

MIT