When you cross-validate, you're usually using stratified sampling. It's a default that comes with scikit-learn, and it certainly has an upside ... but also a downside.
We try to explain why in our latest whiteboard video on YouTube.
www.youtube.com/watch?v=EpK9...
