Bout categorization.

(a) The held-out log likelihood from fitting the Gaussian Mixture Model to the exploratory and aversive datasets vs the number of mixture components. We characterize each bout using its speed, change in heading, summed tail angle (integrated absolute tail angle over time) and the first three PC components of the tail angles. (b) The histograms of speed, change in heading and duration of bout for each bout type.

(EPS)