Neuraldino is an AI for Google's t-rex game. It is written in Javascript and uses the synaptic.js module for creation and utilisation of neural networks. The source code of the game is retrieved from this repository. Feel free to fork this project.
Please note that ducking is not possible in my version of the game, yet
Very simple. Just visit wagenaartje.github.io/neuraldino. Press spacebar to play the game. Once some obstacles in sight you will be adding data to the training sets. Try to have a minimal size of 500, and try not to exceed 5000. Once you are done gathering data, press the Train! button. This will train your sets to the neural network 100 times. Train until you have an error of <0.03. After you're done training, toggle the receiving data button. Now press spacebar, and the AI will take over control.
Does your AI reach a reasonable score? Then export your network, so you can show it off anytime later!
The repository is not very open to adjustment yet. Soon I will be transforming the code to object-oriented design, this will create more overview and thus make editing easier.