Thanks Jason! I have another question: I’ve been looking at your code example (the python notebook) and am a bit puzzled by your method “utils.get_batch”. It seems to be taking a random sample of 500 from the training set to train each time, with the same 100,000 as the test set. Then you pick the best set of parameters from a single 500 training set. Why is this? Can you tell me why you used this method or what it is called?