felix-schott/traffic-accidents

Overview

This is a simple Flask app with a HTML/JS/CSS + Leaflet.js frontend. The app displays the number of traffic accidents per district in Berlin in 2019 relative to area. Users can sample n % of districts and let a pretrained Regression-Kriging model predict the values for the missing districts based on daily traffic volume. The demo can be accessed here.

The folder backend contains all python scripts, including the notebooks cleaning1 and cleaning2 that contain all the data preparation work. Data sources are listed in the readme file in the data directory. train_model contains the actual model. app sets up the API and runs a local development server (the demo is deployed on a linux server using nginx/gunicorn/pm2).

frontend contains the app frontend. All geodata is dynamically requested from the backend (API specified in backend/app.py) and then rendered on the client side using Leaflet.js.

How to run the app on your machine

1. Clone the repo.
2. cd into the backend subdirectory.
3. Create and activate a virtual environment.
4. Run pip install -r requirements.txt to install the dependencies.
5. Run the app with python app.py.

traffic-accidents's Issues

Dockerise/microservices

Description

Containerise the servers, create docker-compose file with postgis db, proxy, frontend and backend server.

Restructure

Description

Restructure code into 4 domains:

• data: contains data ingestion notebooks, push to postgis db
• model: contains the model training notebook
• server: contains the server module
• frontend: stays as is, contains the frontend

Migrate server to FastAPI

Description

Migrate server from Flask to FastAPI.