It seems very odd to recreate the generator with a batch_size slice of the data on each trip through the loop, but that seems to be the desired approach. I get the same result if I create the data_generator with the full test set outside the loop and then just call next(dg) each time through the loop, which seems like the more idiomatic way to use a generator. (It even passes test_classify in spite of my having added a new line of code…)
Am I missing something?
Cheers - Jim