Fig 1. Schematic structures of behavior data in multivariate and multidimensional tensor form.
(A) The schematic structure of a time-series behavior data with one behavior measurement repeatedly collected from multiple samples (blue axis) over time (red axis). The resulting data structure is a matrix. Each brown rectangle represents a unique measurement per sample per time point. Its color intensity represents the magnitude of the measurement. (B) The schematic structure of a time-series behavior data with multiple behavior measurements (green axis) repeatedly collected from multiple samples (blue axis) over time (red axis). The resulting data structure is a three-dimensional array known as a three-dimensional tensor (3DT) or a third-order tensor, a type of multidimensional tensor (MDT). In this array, each “sheet” represents the data collected from one sample. The different color rectangles (brown, green, and blue) represent different behavior measurements per sample per time point, whereas their color intensities represent the magnitude of the measurement. (C) The schematic structure of MDT data when matricized by Feature Concatenation. The matricization process concatenates behavior measurements across a slice of the data cube in (B) at one time point, then iterates through all time points. As a result, the behavior measurement dimension is reduced, transforming the 3DT into a two-dimensional tensor or a matrix.