Fig. 1
Diverse time series (of varying types, sampling rates, and durations) are organized meaningfully in a common feature space. We plot each time series in a reduced, two-dimensional feature space, computed with t-SNE11 from a high-dimensional feature space. Each time series is labeled according to a set of broad categories using color (see Supplementary Information for descriptions of each category). Most categories occupy distinct parts of the space, and categories of data with similar dynamical properties are close in the space. Interesting connections between distinct systems are flagged where distinct classes of overlap; three examples are annotated (as ‘a’, ‘b’, and ‘c’), and some examples of time series in each area are visualized as 1000-sample segments in the right panels of the figure.
