Coding for Fun: Multilayer Perceptron
-
Hi everyone,
I made a little project for machine learning. It uses lists of vectors to try to speed up the math here and there. (i.e., the inputs could be [{1,2,3,4}, {5,4,3,2}] for 8 inputs.
One purpose of machine learning is to make something generate valid outputs from new inputs.
In this case, I used a very small subset of integer numbers, and this perceptron is trained to add/subtract all numbers one frame at a time.
The foundation is intended to handle any number of layers that the memory can handle (and includes backpropagation).
You could use my file system/serialization package to store the weights if you have a more advanced purpose for it (you'll probably change the activation function and the corresponding derivative function to something based on what you are teaching it).
ID: NXKRY6QDN5
Jason
-
Very cool! Will you whip up a demo of this, so one could get an example for how it can be used practically?
-
@chronos I know it would be silly but a simple game like flappy bird might be the perfect example for this. There are not too many inputs and the only outputs are to flap or not.
-
I made it learn to hit a target that moves at different speeds at different heights.
