What is the role of TDD with Machine learning?

I have looked at a good 50 books on machine learning and see few that seriously deal with Testing Driven development and Machine learning.  Given the vary nature of machine learning, the lack of TDD seems to be a problem waiting to happen.  Having grown up in the world of credit risk modeling, I know the level of scrutiny we had at every level of model development deployment and monitoring.  More recently I dived into full-stack web development with a rather simple web app that is backed by 1400 tests in a continuous integration environment.  It is a challenge keeping the environment and test suite in sync at times, but the foundation principles of creating single responsibility classes and overtly spelling out acceptance criteria at the most basic level (unit tests) is a critical discipline.