ugh I had an idea for a neural network layer but I can't think of a way to implement it simply in numpy, nevermind any tensor based libraries. I wish numpy's shuffle function could shuffle rows independently. maybe my best bet is rolling a hashing function or something?