Suppose you want to predict y and you have predictor vectors, each with known accuracy, variance, and sample length. How do you combine these predictors to minimize the residual standard error (RSE)?