ABSTRACT
As a first application of the model based approximation methods of Chapter 14 we consider the problem of estimating themodel elements (R,Q) by sampling. Thismay be based on histories of an active system or on computer simulations. The main problem we consider is the determination of a sample size sufficient to achieve a fixed algorithm tolerance. We assume the concern is entirely with model estimation, but will later consider the problem of determining the optimal rate of exploratory behavior in an active system also subject to optimal cost control (Chapter 18).
