Data scientists often don't have experience working as software developers and never learned how to write unit tests for their code. They may be used to writing and executing code interactively or in an ad hoc fashion; they may be unused to writing code that runs without supervision or as part of a larger pipeline. Things get even more complicated when code has non-deterministic outcomes as is the case with probabalistic models. In this talk, I'll introduce the concept of unit testing, describe how to write good tests, and discuss how to make testing a part of a normal data science workflow. Attendees should be already familiar with the PyData stack but might not be writing production code.