Hi Rob, yes that's correct. The fit is executed in full for each mini batch in this code but can also be performed with a single gradient descent step.