Paraglidable is an A.I.-based flying conditions forecasting program for paragliding.
You can find it live here: https://paraglidable.com

This repository contains:

• Scripts for setting and training the neural network, downloading +10 days forecasts data from third parties and running a prediction in /neural_network/
• Program for generating the map tiles from a prediction in /tiler/
• Complete web site in /www/

The easiest way to start playing with Paraglidable is to use Docker. I will only provide support for this workflow. But you can also check the Dockerfile and install dependencies on your own.

The main dependencies are:

• Python 3
• TensorFlow 2
• Qt 5
• Apache HTTP Server with PHP

git clone https://github.com/AntoineMeler/Paraglidable.git

docker build -t paraglidable Paraglidable/docker/
docker run -it -p 8001:80 -p 8888:8888 -v $(pwd)/Paraglidable:/workspaces/Paraglidable paraglidable

cd /workspaces/Paraglidable/scripts/
python download_data.py             # Download training weather and flights data (200MB)
python download_elevation_tiles.py  # Download elevation data (260MB)
python download_background_tiles.py # Download background tiles (facultative) (180MB)
sh build_tiler.sh                   # Build the C++ tiler

You're all set!

• /neural_network/train.py run a new training
• /neural_network/forecast.py run +10 days forecast and generate tiles
• /scripts/start_server.sh start Apache server to visualize the forecast on the local website

You can find the neural network description here: neural network documentation

Contributions on any subject are welcome by doing a pull request from a fork!