This repository was made to showcase the possibility of running a machine learning model directly in the browser.
See the live demo on my personal website.
NOTE: The municipalities available in the dropdown are currently hard-coded and limited to a few examples. This is due to the fact that this is only a showcase and not a full-fledged application.
The source code is heavily based on KI-Anwendungen-n-to-n-backend and inspired by KI-Anwendungen-n-to-n-frontend by bkuehnis.
The Python code is mostly identical to the backend of the original project, except that no Flask is used. Instead, the model is run with the help of Pyodide, a WebAssembly based distribution of Python. This is done in the background using a Web Worker.
To get the model to work with Pyodide, the web worker fetches the required
bfs_municipality_and_tax_data.csv and randomforest_regression.pkl files
from the server and creates them as new files in the Pyodide environment. The
model is then run with the input data and the result is sent back to the main
thread. The result is then displayed on the website in a similar way to the
original project (can't check due to unavailability of the sample deployment).