Veritas Sandbox

This is a sandbox image of the Veritas diagnosis and assessment tool, and is meant to help financial services assess their AI and Data Analytics models. It wraps both tools into a convenient jupyter lab interface. The docker image consists of three parts:

• Jupyter lab server with Veritas Diagnosis Tool installed
• Example notebooks for Veritas Diagnosis Tool
• Veritas Assessment Tool web application

Installation

The easiest way to get started is to use the docker image. The image is available on Docker Hub.

docker pull cylynx/veritas-sandbox

Create a docker-compose file to start the service. The example below exposes the Jupyter lab server on port 8888 and the assessment tool on port 8001.

version: "3.9"

services:
  veritas-sandbox:
    image: cylynx/veritas-sandbox
    container_name: veritas-sandbox
    ports:
      - 8888:8888
      - 8001:8001

Start the service in detached mode with docker-compose up -d.

Jupyter lab should be available on port 8888 and the assessment tool on port 8001. The Jupyter homepage showcases a few demos including a credit scoring and customer marketing use case. Check out the examples, or create a fresh notebook to test with your own data. It also contains a convenient link to the Assessment tool.

The default username (admin) and password (123456) can be used to access the assessment tool. Try generating a veritas report by uploading the json artifact produced by the python diagnosis tool.

Persist user data

To persist user data, modify the docker-compose setup to use docker volumes. The example below persists the notebook directory in the ./data/veritas folder and the assessment sqlite db and artifacts in ./data/assessment.

version: "3.9"

services:
  veritas-sandbox:
    image: veritastool/veritas-sandbox
    container_name: veritas-sandbox
    ports:
      - 8888:8888
      - 8001:8001
    volumes:
      - type: bind
        source: ./data/veritas
        target: /home/jovyan/veritas
      - type: bind
        source: ./data/assessment
        target: /usr/local/bin/veritas/file

Develop

Clone the repository, run ./build_image.sh to build the custom jupyter lab docker image and docker compose up to start the service.

The custom Jupyter launcher landing page is added as a Jupyter labextension, with the default launcher disabled. To modify the landing page, edit the code in ./veritas-launcher/src and build the labextension with yarn build:prod, before re-building the docker image.

Data and Settings

data/veritas - contains all example notebooks and is the main workspace folder for the user. Added to the built docker image.

settings/jupyter_lab_config.py - Jupyter Lab settings. In the current setup, the default kernel is disabled and the default user directory modified to be consistent with the default user in the docker image.

settings/page_config.json - configuration data to the application environment. The default launcher is disabled, since a custom veritas-launcher is being used. To change the default url path (http://localhost:8001) for the assessment tool, modify veritasAssessmentUrl and rebuild the image.

settings/overrides.json - default value for extension settings. For example, it can be used to change the default theme. Currently, it is not used.

settings/supervisord.conf and settings/run_supervisord.sh - supervisor settings to run both services in a single image. Modify the commands to start the Jupyter or assessment tool services with different default behaviours. For example to launch Jupyter with authentication, the following setting could be used: command=/usr/local/bin/start.sh jupyter lab.

License

Veritas Sandbox is licensed under the Apache License, Version 2.0 - see LICENSE for more details.