Fig. 3

Applying NITPicker. a Monthly temperatures from a set of cities in Canada are shown in black. For the purpose of this paper, we consider the ‘control’ condition to be the temperature of Resolute, Canada, which is shown in red. A probability distribution of curves was constructed on the basis of the temperature curves for all cities – curves sampled from this probability distribution are indicated in dashed-blue lines. The vertical lines represent the ‘best time points’ to sample from, according to NITPicker. b For each of ten replicates, we selected the time points to sample using half the city curves, and scored the selection of time points on the other half of the city curves. For each city in the test set, we calculated the L²-error between the curve generated by sampling every month and the curve generated by linear interpolation between the selected subset of time points. c Here we present an example of how we evaluate a test set for a selection of time points (vertical bars) selected by NITPicker (i), random (ii), and evenly sampled (iii). The coloured-in area displays the error that arrises from sampling only at the designated time points. d The growth rate of boys and girls from the Berkeley growth dataset were used to develop probability distributions of curves for boys and girls, with sampled curves shown in e. f We were interested in estimating the shape of the inverse coefficient of variation, shown in this figure. The selected time points are shown as vertical bars in D-F. g We used half the boy curves and half the girl curves to select time points to sample from, and to train a DD-classifier [18, 19], and then calculated the percent accuracy on the other half of the boy and girl curves. This procedure was repeated 30 times with each method of selecting time points (selecting all the time points, 5 time points with NITPicker, 5 time points randomly, and 5 time points evenly)