Package Overview

A brief introduction to the methods, measures and algorithms provided by pyunicorn.

General complex networks

Many standard complex network measures, network models and algorithms are supported, most of them inherited from the igraph package, e.g., degree, closeness and betweenness centralities, clustering coefficient and transitivity or commmunity detection algorithms and network models such as Erdos-Renyi or Barabasi-Albert. Moreover, a number of less common network statistics like Newman’s or Arenas’ random walk betweenness can be computed. Reading and saving network data from and to many common data formats is possible.

Spatially embedded networks

pyunicorn includes measures and models specifically designed for spatially embedded networks (or simply spatial networks) via the GeoNetwork and Grid classes.

Interacting/multiplex networks (networks of networks)

The InteractingNetworks class provides a rich collection of network measures and models specifically designed for investigating the structure of networks of networks (also called interacting networks, interdependent networks or multiplex networks in different contexts). Examples include the cross-link density of connections between different subnetworks or the cross-shortest path betweenness quantifying the importance of nodes for mediating interactions between different subnetworks. Models of interacting networks allow to assess the degree of organization of the cross-connectivity between subnetworks.

Node-weighted (node-splitting invariant) network measures

Node-weighted networks measures derived following the node-splitting invariance approach are useful for studying systems with nodes representing subsystems of heterogeneous size, weight, area, volume or importance, e.g., nodes representing grid cells of widely different area in climate networks or voxels of differing volume in functional brain networks. pyunicorn provides node-weighted variants of most standard and non-standard measures for networks as well as interacting networks.

(Coupled) Climate networks

pyunicorn provides classes for the easy construction and analysis of the statistical interdependency structure within and between fields of time series (functional networks) using various similarity measures such as Pearson and Spearman correlation, lagged linear correlation, mutual information and event synchronization. Climate networks allow the analysis of single fields of time series, whereas coupled climate networks focus on studying the interrelationships between two fields of time series. While there is a historical focus on applications to climate data, those methods can also be applied to other sources of time series data such as neuroscientific (e.g., FMRI and EEG data) or financial data (e.g., stock market indices).

Recurrence quantification/network analysis

Recurrence analysis is a powerful method for studying nonlinear systems, particularly based on univariate and multivariate time series data. Recurrence quantification analysis (RQA) and recurrence network analysis (RNA) allow to classify different dynamical regimes in time series and to detect regime shifts, dynamical transitions or tipping points, among many other applications. Bivariate methods such as joint recurrence plots/networks, cross recurrence plots or inter system recurrence networks allow to investigate the coupling structure between two dynamical systems based on time series, including methods to detect the directionality of coupling. Recurrence analysis is applicable to general time series data from many fields such as climatology, paleoclimatology, medicine, neuroscience or economics.

Visibility graph analysis

Visibility graph analysis is an alternative approach to nonlinear time series analysis, allowing to study among others fractal properties and long-term memory in time series. As a special feature, pyunicorn provides time-directed measures such as advanced and retarded degree/clustering that can be used for designing tests for time-irreversibility (time-reversal asymmetry) of processes.

Surrogate time series

Surrogate time series are useful for testing hypothesis on observed time series properties, e.g., on what features of a time series are expected to arise with high probability for randomized time series with the same autocorrelation structure. pyunicorn can be used to generate various types of time series surrogates, including white noise surrogates, Fourier surrogates, amplitude adjusted Fourier (AAFT) surrogates or twin surrogates (conserving the recurrence structure of the underlying time series).