Neural Network
ASCII Game
Here is a self-trained neural network navigating obstacles. 300 randomized neural networks were set to play the game and graded based on how far they got. There were other minor factors such as penalties for moving more to keep its movement asthetically pleasing for us to watch. The 'best' 100 were chosen to move on to the next generation. The 100 themselves were carried on directly, and to get back up to the original 300, 200 more were generated by 'genetically' mixing the 100 parents with a small amount of random mutation. The neural net you see playing the ASCII game below is the highest rated after 1000 generations.
It is fed the three rows in front of it from its perspective as input, and has an output layer of size 3, corresponding to up, don't move, and down.
Here is the serialized neural network that I saved and loaded to record it playing above.
