Given a set of graph statistics $S = (s_1 , s_2 , ..., s_n)$ (these statistics can include degree statistics, clustering coefficients, or motifs or graphlets), compute the each statistic $s_i$ for both the generated graphs and a test graph. From there, we can compute the distance between the statistics' distributions on the test graph and generated graph using a distributional measure, such as the total variation distance:

$d(s_{i,G_{test}}, s_{i,G_{gen}}) = \sup_{x \in \mathbb{R}} |s_{i,G_{test}}(x) - s_{i,G_{gen}}(x)|$

Finally, we can compute the average pairwise distributional distance between a set of generated graphs and graphs in a test set for each statistic $s_i \in S$