This was just an experiment born out of my curiosity about neural networks. This is a genetic neural network that can teach itself how to play flappy bird. I'm not in high school yet, so I didn't incorporate backpropagation (as that requires some calculus), but someday I might add it. For those who don't know, a neural network is an AI system that allows computers to learn almost anything. This neural network includes 4 input nodes (the bird's y velocity, the birds y position, the distance to the pipe, and the pipe's y position) which are each multiplied by 4 weights. A weight is just a number for the input nodes to be multiplied by. We then add up node1*weight1, node2*weight2, node3*weight3, and node4*weight4. This value is then plugged into an altered sigmoid function which basically means that it is converted into a number between -1 and 1. The bigger the number, the closer it is to 1 and the smaller the number, the closer it is to -1. This final value is the first node of the next layer. This process is repeated for the other nodes. This process is repeated again, but instead of using the input nodes, you use the nodes of the second layer. This time, there is only 1 node in the next layer rather than 4. This 1 node is called the output node. If it ends up bigger than 0, the bird flaps, if not, it doesn't. If this didn't make sense, or if you still have questions, just leave a comment, or watch 3blue1brown's YouTube series on the subject.
Code 100% me. @CL_Games made a similar project to this at https://scratch.mit.edu/projects/538068927/ so shoutout to them. This was not copied. Use turbomode to train faster.
