In Supervised Machine Learning, in Regression Algorithms, after training a model, how do you know if it was efficient? Regarding the mean square error, the better the smaller, but what about the training-test data, what relationship should they have and what does that mean?