authenticity-portal

This is a prototype of a web site/server (was hosted under https://authenticity-portal.herokuapp.com/ until Heroku stopped its free tier) which

• serves selected 'prediction models'* to a general audience
• presents the models within a specific topic (use case), providing additional information and context

It is named authenticity-portal to indicate one possible field of application: A platform where one can upload some (measurement) data of a food to verify/predict certain properties, like authenticity of the geographical organic. The prediction models are curated and validated by trusted institutions.

(*note that in my terminology, prediction models include classification)

Design description

Targeted use case

The portal is designed to accomodate two different work flows (and user types):

• The general user uses validated models to obtain information about their dataset. For example:

  • a quantitative prediction, like the nutrient content of a sample
  • a classification, like the authenticity of a sample (likelihood of class member ship) This user typically has limited access, the main interaction with the portal being:
  • access details/information about the selected topics and the available models
  • request a prediction by feeding a data file to one of the hosted models

• The data manager or scientist curates the validated models. This involves

  • upload of measurment data
  • generation of models
  • training of models on selæected measurements
  • validation/staging a model for use by general users This user typically has advanced access to the site. which might include access to the backing data.

Implementation highlights

This is my first web site, where I focussed mainly on the web server part and much less on the frontend.

One highlight is the custom port of the Simca model from R (mdatools) to python, thereby providing an interesting use case that goes beyond usual calssification models.

Technical tools

• Backend / web server framework: django (python)
• Frontend framework: Bootstrap 5.0
• Configured for deplyment on the heroku platform

Deployment instructions

The subfolder portalsite is setup for deployment to heroku (with a Postgres db), for example by the git command

git subtree push --prefix portalsite heroku main

and contains therefore i.p. the files

• requirements.txt
• runtime.txt
• Procfile

Note that the target environment is expected to provide the variables 'DATABASE_URL', 'DJANGO_DEBUG' and 'DJANGO_SECRET_KEY'. And if required, a superuser should be created after intial deployment, see heroku and django docs for details.

The database can be seeded via heroku CLI with the command

heroku run python manage.py loaddata portal/fixtures/initial_seed_data.json

This will create iris-example data with two models (trained on Setosa) as well as two users

• 'generic_test_user', password 'GENER1C!' with only basic rights
• 'scientist_test_user', password 'SC1ENCE!' with rights to all portal objects and selected administration privileges