Figure 2
From: Temporally aligned segmentation and clustering (TASC) framework for behavior time series analysis
TASC framework flow. (A) Multidimensional continuous behavior time series. (B) Candidate periods of activity identified following normalization. (C) Initial segmentation. (D) Embedding in latent space followed by (E) clustering. (F) For each segment, variants are generated from its temporal neighbors and are scored relative to its cluster’s center using linear alignment. The variant is “shifted” and “warped” such that the distance between the aligned version and the center is minimal, which is equivalent to the maximal coincidence of the sequences. Two example variants are shown, for each one the dashed lines mark the original neighbor, and the solid lines—the aligned version. (G) Elimination of overlaps and filling of significant gaps for highly scoring segments. Dashed lines mark overlapping candidate segments, solid—the best choice after overlap elimination. Green and red lines mark the beginning and end of segments respectively.
