Deep scattering transform on segmented time series.
This program contains a series of command-line tools for clustering continuous time series with a deep scattering network. The following sub- commands must be run in a specific order from the continuous data to the cluster results.
-
The inventorize command lists and selects the data based on usual meta parameters (sampling rate, duration, channels) and stores the relevant datapaths into an inventory file. This first command helps explore the data coverage in time, selecting appropriate time segments, and running the remaining steps on the actual data.
-
The transform command runs the deep scattering transform on the segmented time series and stores the scattering coefficients for later feature extraction.
-
With the features command, the large-dimensional scattering coefficients are reduced to a low-dimensional space which dimensions are considered features here. These features are used in the clustering step next.
-
The command calculates the linkage matrix that helps cluster the data based on some criteria of similarity (a metric and a method). Once this matrix is calculated, the clusters are extracted for further analyses.
Created in May 2021 by Leonard Seydoux and Randall Balestriero.