Effective testing for machine learning systems.

Working as a core maintainer for PyTorch Lightning, I've grown a strong appreciation for the value of tests in software development. As I've been spinning up a new project at work, I've been spending a fair amount of time thinking about how we should test machine learning systems. A couple weeks ago, one of my coworkers sent me a fascinating paper on the topic which inspired me to dig in, collect my thoughts, and write this blog post. In this blog post, we'll cover what testing looks like for traditional software development, why testing machine learning systems can be different, and discuss some strategies for writing effective tests for machine learning systems. We'll also clarify the distinction between the closely related roles of evaluation and testing as part of the model development process.