Watch: Overfitting vs Generalizing

You just fit a model to noisy data. The big question in machine learning is coming up: how flexible should a model be? Too rigid and it misses the pattern; too flexible and it memorizes the noise instead of learning the trend.

Press play. Watch a simple model follow the true pattern, then watch a complex model contort to hit every training point, zero training error, and see why that is exactly the wrong thing on new data. Then you will code the underfitting/overfitting tradeoff yourself.