Figure 1.

Transformation of movement time series into a Symbolic Movement Representation. Schematics of the pipeline: raw data from wrist sensor is first subsampled, band-passed filtered, and taken in windows of 1 second with overlap of 240 ms. Each window is mapped to a discrete syllables (Clustering-Tokenization). By associating each window of activity to a cluster, the continuous signal is transformed into a discrete sequence of syllables (Sequence-Embedding). The sequence can be seen as a discrete markov chain and each action can be then represented by a Transition Matrix from a Markov-Chain. An SMR (see main text) is estimated as the Symbolic Distribution estimator of the transition matrix obtained performing several actions.