When we need to predict more than just a mean or a median, full posterior distributions from Bayesian models are often the way to go. But sometimes, that’s too computationally intensive and we need some shortcuts. Quantile regression is a handy alternative. For even more efficiency, we can use multi-task learning so that a single model produces all the quantiles we want.
As we add outcomes to our model, the concepts stay the same but the dynamics grow more complex. Viewing animations of the model can help us develop intuitions about how it works.
