Lawrence Livermore National Laboratory conducts multiscale scientific simulations to answer questions where experiments are infeasible. To reduce computational costs, accurate yet expensive computations can be approximated using neural networks. To this end, it is important to know how much data is needed to achieve specified accuracy. Previous research shows that neural networks obey empirical scaling laws. Using this result, we developed software that measures how accuracy scales as a function of data and network size for each specific scientific domain.