Playing with #deeplearning / #nlp with #Python and #Pytorch.

It's sort of annoying that sometimes my code would fail with some type error or dimension mismatch after running for tens of minutes. Currently dealing with this by testing on a smaller dataset first after code changes, but it still feels a bit unwieldy.

Also, my preprocessing code takes more time than the inference itself... Wonder if there's a good #profiling tool for Python.