We develop sublinear-time algorithms for computing and optimizing opinion-based quantities in the FJ model using partial rooted forest sampling.