So I recently tested the results of stages in my inversions and found that at 5k steps there is very little variance so I have have my training steps. so now the longest part of my process is sorting the data set. So I have done this one from 4 sets and just let the pre-process do all the work.