Air Quality Development and Model Selection

The outcome of this project can be seen at airquality.sneezingtrees.com - an online tool for predicting air quality in Montpellier. I recommend you start there as most of the initial background and explanatory information is available there.

This is one of two repositories that go hand in hand:

• Air Quality Development and Selection - contains all work done to develop and select a machine learning model for predicting the concentrations of the five pollutants used to measure European Air Quality.
• Air Quality Prediction Application (where you are now) - the code for the web application at airquality.sneezingtrees.com

Data

All the data required for this project that is not available in this repository (and its sibling) is available on Google Drive. The "live" data that the website uses is as at 21 October 2022. You will need to request an API key from Geod'Air to update beyond that (it's free).

Local deployment

• Make sure you have a .env file in the root directory i.e. the parent directory of the app and update_app folders. Use the .env-sample file to help you.
• Download the app_data folder from Google Drive and put it somewhere in your filesystem (where you as a user have access). Set the WORK_DIR and BIND_MOUNT_PATH values in your .env file to that path.

Flask / Gunicorn

Flask

Run flask --app air_quality_prediction --debug run from the app folder. (You don't have to use the --debug option but it can be very helpful.) The site will then be available at http://localhost:5000/.

Gunicorn

Run gunicorn -w 4 --bind 0.0.0.0:8000 wsgi:app. (The -w option is the number of workers. Adjust to suit your computer.) The site be available at http://localhost:8000/.

Docker

Run with the command docker compose up in a terminal in the parent directory of the app and update_app folders where the docker-compose.yml file is.

Cloud hosted

1. Create an app_data folder on the host and upload all the data in the app_data folder on Google drive. Set the folder and all of its contents to permssions 777.
2. Upload a docker version of the .env file (i.e. WORK_DIR key-value pair is excluded) to the app_data folder on the host.
3. Upload the file docker-compose cloud.yml to the app_data folder on the host and rename the file to docker-compose.yml.
4. Run build command docker build -t {your_docker_hub_username}/aq-prediction:latest --build-arg CACHEBUST=$(date +%s) . in app folder on your local machine.
5. Run build command docker build -t {your_docker_hub_username}/aq-update:latest --build-arg CACHEBUST=$(date +%s) . in update_app folder on your local machine.
6. Push to docker hub from your local machine with docker push {your_docker_hub_username}/aq-prediction:latest and docker push {your_docker_hub_username}/aq-update:latest.
7. Log in to your host via SSH and navigate to the app_data folder. Run docker compose up --detach command.