Problem of sampling

We can estimate policy $\pi_{2}$ from samples of another policy $\pi_{1}$. The problem is the variance can be big -infinte. (variance in example 1). Big variance can cause the error of the estimation grow.