ASDF-Dashboard

This web application provides a service for an automated subgroup fairness analysis of a binary classifier. Our system detects the subgroups in the data automatically by using either subgroups obtained from clustering the dataset or entropy-based patterns derived from the found clusters.

Therefore, the ASDF-Dashboard provides an easy automatic computation of the subgroups with the SLINK clustering algorithm. Alternatively, a customizable fairness assessment supporting different, configurable clustering algorithms is offered, too. The results of the automatic subgroup fairness analysis of the binary classifier are visualized in the dashboard. The interactive charts provide various perspectives and deep insights into the classifier's behavior on the detected subgroups.

We offer a live demo of the ASDF-Dashboard service. The ASDF-Dashboard can also be self-hosted. The installation and deployment instructions for the self-hosted setup are given below.

For further information on the scientific details and the experimental results see our ADBIS '22 paper: Clustering-based Subgroup Detection for Automated Fairness Analysis (DOI: 10.1007/978-3-031-15743-1_5).

Self-hosted setup

The easiest way to self host the application is to build and run the provided docker compose setup consisting of

• a nginx service providing the reverse proxy functionality,
• a celery worker service for the distribution of the background tasks,
• a redis database service used for caching and as background task backend,
• a postgresql database service for user and dataset management,
• and the flask container hosting the ASDF-Dashboard service.

Create .env file

Create a .env-file in the project's root folder with the following variables and replace the <placeholders> with actual values according to your environment (template for the .env-file).

# Mail
MAIL_SERVER=<url_smtp_server>
MAIL_PORT=<port_smtp_server>
MAIL_USERNAME=<user_smtp_server>
MAIL_PASSWORD=<password_smtp_server>
MAIL_USE_TLS=<flag_tls_smtp_server(True/False)>
MAIL_SENDER=<sender_email_address>

# General
COMPOSE_PROJECT_NAME=asdf-dashboard
SECRET_KEY=<some_secret_key>
SALT=<some_salt>

# Docker
DOCKER_RESTART_POLICY=unless-stopped
DOCKER_STOP_GRACE_PERIOD=3s
DOCKER_WEB_PORT=8000
DOCKER_WEB_VOLUME=./instance:/app/instance

# Flask
FLASK_ENV=production

# Nginx
NGINX_PORT=<nginx_port>

# Gunicorn
WEB_RELOAD=<flag_web_reload(true/false)>
WEB_CONCURRENCY=<number_of_workers>

# Postgres
POSTGRES_DB=<db_postgres>
POSTGRES_USER=<user_postgres>
POSTGRES_PASSWORD=<password_postgres>

Configure Nginx

Adapt the nginx configuration template to your environment. To this end, copy it into a file named nginx.conf, which is located at the same folder. The listening port and the server name have to be set according to your host/domain, that will be pointed to the web service, and the same value as the NGINX_PORT environment variable (in the .env-file).

...
server {
        listen {same_as_the_NGINX_PORT};
        server_name {your_domain_name};
        ...
}
...

If you want to set a different user disk quota, you have to set the MAX_QUOTA_MB environment variable in the .env-file and configure limit for the client body size (client_max_body_size) in the nginx configuration, e.g.,

...
http {
    client_max_body_size 15M;
    ...
}

The number of worker processes (worker_processes) can also be increased. It is not recommended to change other values from the nginx configuration template.

Deployment under a url prefix

If you want to deploy the ASDF-Dashboard app under an url prefix (e.g., /asdf), use the nginx configuration template with an url prefix as template for the generation of the nginx configuration file. Also, set the APPLICATION_ROOT variable in the config.py accordingly.

Update footer

Update the footer.html accordingly to tell the users who is hosting this dashboard.

Production setup

Build and run the docker container setup with

docker-compose up --build

After the successful build and start of the containers, the web service of the ASDF-Dashboard should be available via a web browser under http://yourdomain:{nginx_port} as configured in the .env-file, e.g., under http://example.com:80/. This information is required for the ASDF-Dashboard application when sending out account confirmation emails containing confirmation links to the registered users.

Development setup

For a development setup it is sufficient to have a

1. a local redis instance (e.g., in a docker container as shown below),
2. a celery worker instance,
3. and the ASDF-Dashboard Flask application.

Note: The development setup should not be used in a production environment for security reasons.

Install requirements

Install the required python packages with

pip install -r requirements.txt

Redis docker setup

First pull the redis docker image

docker pull redis

and then run it in a container

docker run --name redis -p 6379:6379 -d redis

Celery worker

From the project root execute

celery -A app.celery_app worker -l info

Note: The command line parameter -P has to be set to solo on windows machines.

celery -A app.celery_app worker -P solo -l in

Flask Application

The flask application for the development/debug setup is started by executing the following command:

python app/__init__.py