Classification results.

This table contrasts the classification results obtained through generative embedding with those afforded by three conventional methods. As described in the main text, the underlying dataset serves illustrative purposes, and so, due to its small sample size (n = 37), all numbers are associated with considerable uncertainty. The measure of primary interest is the balanced accuracy (2). Its uncertainty can be captured by computing a posterior probability interval (as shown in Figure 6a), or by computing a p-value (3), which represents the probability with which the observed performance would have been obtained by chance.