Skip to main content
. 2012 Aug 13;109(35):14259-14264. doi: 10.1073/pnas.1203021109

Fig. 3.

Fig. 3.

Fight size distribution for the three generative models. The sparse coding model performs best with the smallest Kullback-Leibler divergence DKL from the true distribution. Here we use 105 fight samples from each model where fights with one or zero participants have been omitted. Shaded areas connect error bars indicating bootstrapped 95% confidence intervals.