Description
This is a great start - now let's move outside of the IPython notebook format, and break this code up into individual .py files. For now start by creating a new branch called mnist-single-file where we're just going to manually turn the entire notebook into a single mnist.py file than you can run via python mnist.py.
After you're done on that branch, make a Pull Request, to merge it back into this repository, and assign me to review. After that, we're going to "break apart" the file based on functionality/reusability - make separate files/directories for the various components that go into training a model.
The hope is that if you want to use the MNIST dataset with a different model, factoring the code this way could help. Similarly if you want to use the same model with a different dataset, or part of the model for a different task - introducing new abstractions is useful!