I have a trained GAN that generates synthetic flight data (9 state variables/features). I plotted the correlation for each pair of state variables similiar to Seaborn's pairplot(). But I have two sets of correlations for each pair: one set for the real data (the blue data in the plots), and the second set for the generated data (the red data in the plots). What is a good metric I can use to evaluate how well the red data is covering the blue data? The ideal case is that the red data cluster is perfectly covering the blue data. I have looked at JS and KL divergences, but they are for probability distributions, not data distributions. Both sets of data are 720000 by 9.