see the early history of learning, it was a discovery that is actually counter intuitive and a common trap for beginners. (don't minimize the training error) I

see the early history of learning, it was a discovery that is actually counter intuitive and a common trap for beginners. (don't minimize the training error) I have seen new PhDs read about it "in theory", but not internalise it for practice, and then they go off an do Bayesian structure learning without a validation set. This DOES happen. This post is to hammer into the brains of any beginner thinking about machine learning that understanding the validation set's purpose is the most important thing to internalise first. e.g. Machine learning is easier than it looks: https://news.ycombinator.com/item?id=6770785 https://news.ycombinator.com/item?id=6770785