So, about 17,300 spectographs were trainin on the Xavier for about 10 hours. I then fed 177 'unseen' spectographs into the network with the saved model and each result looked pretty much as above. There were 7 false positives, which was expected, and no obvious signs of over fitting.