Extended Data Fig. 3: Comparison of decoder performance using rest vs. neutral moments.

Decoders were trained using rest instances vs, positive (blue) and negative (red) affective behaviours. All panels are comparing these decoders with the neutral vs. affective behaviours as shown in Fig. 2. Green and orange curves show the original model AUCs for positive and negative decoders, respectively. The boxplots show the sample distribution of the average AUC for positive behaviours vs. neutral (green, n = 10 participants), and positive behaviours vs. rest (blue, n = 9 participants) in the top row and negative behaviours vs. neutral (orange, n = 5 participants) as well as negative behaviours vs. rest (red, n = 5 participants) in the bottom row. There were no significance difference between the positive (p = 0.6, two-sided non-parametric pairwise ranskum test) and negative (p = 0.66, two-sided pairwise ranksum test) decoders. In the box plots, central lines represent the median and the two edges represent 25 and 75 percentiles; whiskers show the most extreme datapoints, and outliers are shown individually (see MATLAB boxplot function).