Combining the predictors next up previous
Next: An improved model for Up: Trace-driven simulator Previous: Results: Predictor B

Combining the predictors

Since the two predictors perform well under different circumstances, it is natural to combine them by using each when it is most accurate. Comparing the values of CC for various ranges of n', we find that Predictor A does best for small values of n' while Predictor B does best for larger values. Using Predictor A when n' is less than 32 and Predictor B when it is greater, we get a combined predictor that is more accurate than either predictor alone (CC = 0.65).



Allen Downey
Fri May 30 15:09:42 PDT 1997