Package for network sparsification.

Description

An R package for network sparsification with a variety of novel and known network sparsification techniques. All network sparsification reduce the number of edges, not the number of nodes. A network is usually a large, complex weighted graph obtained from real-world data. It is commonly stored as an adjacency matrix or edge list. Network sparsification is sometimes referred to as network dimensionality reduction.

Getting Started

Install and load devtools package:

install.packages("devtools")

Use install_github function to pull and install simplifyNet in your session:

install_github("kramera3/simplifyNet")

Prerequisites

The following packages are required:

igraph, sanic, Matrix, tidyr, methods, fields, stats, dplyr

Also set up the working directory:

setwd("<em>working directory</em>")

simplifyNet

simplifyNet is a R package for network sparsification. It contains a suite of different network sparsification algorithms to output a sparsified network.

Global Network Sparsification:

Global network sparsification. Uses a threshold cutoff to remove all edges below a certain edge weight or removes a certain proportion of lowest edge weight edges.

gns(E_List, remove.prop, cutoff)

Arguments

• E_List: An edge list of the format |n1|n2|weight|)
• remove.prop: The proportion of highest weighted edges to retain. A value between 0 and 1.
• cutoff: Threshold value for edge weight thresholding.

LANS:

Local Adaptive Network Sparsification from the paper by Foti et al.

lans(Adj, remove.prop, output)

Arguments

• Adj: Weighted adjacency matrix of the network.
• remove.prop: Alpha value threshold to designate statistically "unimportant" edges by edge weight.
• output: Designates if the output should be directed or undirected. Default is that the output is the same as the input based on adjacency matrix symmetry. If the default value is to be overridden, set as either "undirected" or "directed".

Sparsification by Edge Effective Resistances:

Sparsification by sampling edges proportional to their effective resistances as formulated by Spielman and Srivastava. This requires two discrete steps: (1) approximating the effective resistances for all edges, (2) sampling them according to the method devised by Spielman and Srivastava.

effR = EffR(E_List, epsilon, type="kts", tol)
EffRSparse(n, E_List, q, effR)

1. EffR, effective resistances calculator.

   • E_List: Edge list formatted | n1 | n2 | weight |.

   • epsilon: Governs the relative fidelity of the approximation methods 'spl' and 'kts'. The smaller the value, the greater the fidelity of the approximation and the greater the space and time requirements. Default value is 0.1.

   • type: There are three methods.