Onion clustering

Code for single-point clustering of time-correlated data.

Dynsight acts as an interface to the clustering algorithm implemented at https://github.com/matteobecchi/onion_clustering and described in this paper ( https://doi.org/10.1073/pnas.2403771121).

Usage

There are two different implementations of Onion Clustering. The most recent, which is called with these functions and classes:

• onion_uni_smooth
• OnionUniSmooth
• onion_multi_smooth
• OnionMultiSmooth

avoids the signals’ segmentation which was present in the original implementation (the one described in the PNAS paper). With this version, the clusters are guaranteed to be stable for at least ∆t frames, but there is no segmentation of the time-series in consecutive windows. This avoids the excessive growth of unclassified data points when increasing ∆t, and helps to have a more precise estimate of the physical timescales within the time-series.

The original implementation, instead, uses the segmentation of the time-series into consecutive windows of length ∆t, which are then clustered individially. It is called with these functions and classes:

• onion_uni
• OnionUni
• onion_multi
• OnionMulti

Example code for performing clustering on time-series can be found in examples/ onion_uni.py and examples/onion_multi.py, for univariate and multivariate time-series respectively. Refer to https://github.com/matteobecchi/ onion_clustering for a complete documentation.