Machine Learning Weather Visualisations
This project attempts to create a living program that visualises live weather conditions from around the world. The piece consists of three parts, a Processing program that acts as a feature extractor using the OpenWeatherMap API, a Wekinator project which trains and runs the machine learning model, and a flexible Processing sketch with many parameters which acts as an abstract visualisation of the incoming data. The visualisation sketch is built around an object-oriented particle-system approach to graphics, where each ‘particle’ is a semi-random collection of vertices connected to make a curly bezier curve. Other parameters available to the machine learning process include rotation, colour settings, movement behaviours and shape characteristics. The aim is for the program to create very abstract representations of the incoming weather data. By training a machine learning model on a series of example cities, a new user can explore additional cities to see how the program visualises them.