This visualization shows how a predictive model improves as data is added to it. The blue dots represent the true values, in this case my electricity usage each day for a year, and the red dots represent the predicted value for that day based on temperature and time.
The test set is the entire dataset. The training set is a subset of the test set (I know that's stupid, but it makes for a cool visualization). This visualization starts out where the training set is only the first few days, and iterates all the way until the training set and test set are the same.
PS, I know that a trivial model could match each day perfectly once all of the data is added; after all, the model "knows" each true value. That sort of misses the point, insofar as there is a point at all.