Dynamically changing network of inter personal communication among the visitors of a beach in southern California.

Description

A data set containing the dynamic network of inter personal interactions among the visitors of a beach in southern California.

Usage

beach

Format

A list with 31 elements, each element represent one observation in time. Each element is a network of varying size.

Source

Almquist, Z. W. and C. T. Butts (2014b). Logistic network regression for scalable analysis of networks with joint edge/ vertex dynamics. Sociological Methodology 44 (1), 1-33.

binaryPlot

Description

Plot for binary matrices, especially adjacency matrices.

Usage

binaryPlot(x, axlabs = TRUE, ...)

Arguments

Details

binaryPlot

clustCoef

Description

Calculates the cluster coefficient from a network

Usage

clustCoef(x)

Arguments

Details

Given a network in the form of adjacency matrix, this calculates the cluster coefficient. For a definition of cluster coefficient, please refer to the igraph documentation.

Value

scaler

Author(s)

Abhirup

Examples

clustCoef(beach[[1]][, ])

dnr: A package for simulating dynamic networks using ERGM family models.

Description

This package provides functions for fitting lagged exponential family models on dynamic network data, simulation from the models and model diagnostics.

note

This package was developed with help from ARO YIP award \#W911NF-14-1-0577.

references

Abhirup Mallik & Zack W. Almquist (2019) Stable Multiple Time Step Simulation/Prediction From Lagged Dynamic Network Regression Models, Journal of Computational and Graphical Statistics, 28:4, 967-979, <DOI: 10.1080/10618600.2019.1594834>.
Abhirup Mallik and Zack W. Almquist (2017). "An R Package for Dynamic Network Regression." Working paper. University of Minnesota.
Zack W. Almquist and Carter T. Butts (forthcoming). "Dynamic Network Analysis with Missing Data: Theory and Methods." Statistica Sinica. <doi:10.5705/ss.202016.0108>.
Zack W. Almquist and Carter T. Butts. (2013). "Dynamic Network Logistic Regression: A Logistic Choice Analysis of Inter- and Intra-group Blog Citation Dynamics in the 2004 US Presidential Election." Political Analysis, 21(4), 430-448.
Zack W. Almquist and Carter T. Butts. (2014). "Bayesian Analysis of Dynamic Network Regression with Joint Edge/Vertex Dynamics." In Bayesian Inference in the Social Sciences. Ed. by I. Jeliazkov and X.-S. Yang. Hoboken, New Jersey: John Wiley & Sons.

Implementation of simulation engine for dynamic networks using smoothing estimates of change statistics.

Description

Implementation of simulation engine for dynamic networks using smoothing estimates of change statistics.

Usage

engineEdge(
  start_network,
  inputcoeff,
  ns,
  model.terms,
  model.formula,
  graph_mode,
  group,
  intercept,
  exvar,
  maxlag,
  lagmat,
  ylag,
  lambda = NA,
  method = "bayesglm",
  alpha.glmnet,
  paramout = TRUE
)

Arguments

Value

list: out_network: list of predicted networks coefmat: if paramout is TRUE, matrix of coefficients at all time.

Author(s)

Abhirup

Examples

## Not run: 
input_network=rdNets[1:6];
model.terms=c("triadcensus.003", "triadcensus.012", "triadcensus.102", "triadcensus.021D", "gwesp");
model.formula = net~triadcensus(0:3)+gwesp(decay = 0, fixed=FALSE, cutoff=30)-1;
graph_mode='digraph';
group='dnc';
alpha.glmnet=1
directed=TRUE;
method <- 'bayesglm'
maxlag <- 3
lambda=NA
intercept = c("edges")
cdim <- length(model.terms)
lagmat <- matrix(sample(c(0,1),(maxlag+1)*cdim,replace = TRUE),ncol = cdim)
ylag <- rep(1,maxlag)
lagmat[1,] <- rep(0,ncol(lagmat))
out <- paramEdge(input_network,model.terms, model.formula,
                graph_mode="digraph",group,intercept = c("edges"),exvar=NA,
                maxlag = 3,
                lagmat = lagmat,
                ylag = rep(1,maxlag),
                lambda = NA, method='bayesglm',
                alpha.glmnet=1)
#
start_network <- input_network
inputcoeff <- out$coef$coef
nvertex <- 47
ns <- 10
exvar <- NA
tmp <- suppressWarnings(engineEdge(start_network=start_network,inputcoeff=inputcoeff,ns=ns,
                     model.terms=model.terms, model.formula=model.formula,
                     graph_mode=graph_mode,group=group,intercept=intercept,
                     exvar=exvar,
                     maxlag=maxlag,
                     lagmat=lagmat,
                     ylag=ylag,
                     lambda = NA, method='bayesglm',
                     alpha.glmnet=alpha.glmnet))
## End(Not run)

Implementation of simulation engine for dynamic networks using smoothing estimates of change statistics.

Description

Implementation of simulation engine for dynamic networks using smoothing estimates of change statistics.

Usage

engineEdgeBayes(
  start_network,
  inputcoeff,
  ns,
  model.terms,
  model.formula,
  graph_mode,
  group,
  intercept,
  exvar,
  maxlag,
  lagmat,
  ylag,
  lambda = NA,
  method = "bayesglm",
  alpha.glmnet,
  paramout = TRUE,
  Theta = NA
)

Arguments

Examples

## Not run: 
startNet <- rdNets[1:50]
model.terms=c("triadcensus.003", "triadcensus.012", "triadcensus.102", "triadcensus.021D", "gwesp")
model.formula = net~triadcensus(0:3)+gwesp(alpha=0, fixed=FALSE, cutoff=30)-1
graph_mode <- 'digraph'
group <- 'dnc'
alpha.glmnet <- 1
method <- 'bayesglm'
maxlag <- 3
lambda <- NA
intercept <- "edges"
cdim <- length(model.terms)
lagmat <- matrix(sample(c(0,1),(maxlag+1)*cdim,replace = TRUE),ncol = cdim)
ylag <- rep(1,maxlag)
lagmat[1,] <- rep(0,ncol(lagmat))

out.coef <- paramEdge(input_network = startNet,
                model.terms = model.terms,
                model.formula = model.formula,
                graph_mode='digraph',
                group=group,intercept = intercept,
                exvar=NA,
                maxlag = maxlag,
                lagmat = lagmat,
                ylag = ylag,
                lambda = NA, method='bayesglm',
                alpha.glmnet=1)


inputcoeff <- out.coef$coef$coef.edge
nvertex <- 47 ##find vertex here
ns <- 1
exvar <- NA
for(i in seq_along(startNet)) Theta <- Theta + startNet[[i]][,]
Theta <- Theta/length(startNet)
Theta <- thresh(Theta)
out.bayes <- engineEdgeBayes(start_network=startNet,
inputcoeff=inputcoeff,
ns=ns,
model.terms=model.terms,
model.formula=model.formula,
graph_mode=graph_mode,
group=group,intercept=intercept,
exvar=exvar,
maxlag=maxlag,
lagmat=lagmat,
ylag=ylag,
lambda = NA, method='bayesglm',
alpha.glmnet=alpha.glmnet,
Theta = Theta)

## End(Not run)

Implementation of simulation engine for dynamic networks without using smoothing estimates of change statistics.

Description

Implementation of simulation engine for dynamic networks without using smoothing estimates of change statistics.

Usage

engineEdgeNS(
  start_network,
  inputcoeff,
  ns,
  model.terms,
  model.formula,
  graph_mode,
  group,
  intercept,
  exvar,
  maxlag,
  lagmat,
  ylag,
  lambda = NA,
  method = "bayesglm",
  alpha.glmnet,
  paramout = TRUE
)

Arguments

Value

list: out_network: list of predicted networks coefmat: if paramout is TRUE, matrix of coefficients at all time.

Author(s)

Abhirup

Examples

## Not run: 
input_network=rdNets[1:6];
model.terms=c("triadcensus.003", "triadcensus.012", "triadcensus.102", "triadcensus.021D", "gwesp");
model.formula = net~triadcensus(0:3)+gwesp(decay=0, fixed=FALSE, cutoff=30)-1;
graph_mode='digraph';
group='dnc';
alpha.glmnet=1
directed=TRUE;
method <- 'bayesglm'
maxlag <- 3
lambda=NA
intercept = c("edges")
cdim <- length(model.terms)
lagmat <- matrix(sample(c(0,1),(maxlag+1)*cdim,replace = TRUE),ncol = cdim)
ylag <- rep(1,maxlag)
lagmat[1,] <- rep(0,ncol(lagmat))
out <- paramEdge(input_network,model.terms, model.formula,
                graph_mode="digraph",group,intercept = c("edges"),exvar=NA,
                maxlag = 3,
                lagmat = lagmat,
                ylag = rep(1,maxlag),
                lambda = NA, method='bayesglm',
                alpha.glmnet=1)
#
start_network <- input_network
inputcoeff <- out$coef$coef
nvertex <- 47
ns <- 10
exvar <- NA
tmp <- suppressWarnings(engineEdgeNS(start_network=start_network,
                     inputcoeff=inputcoeff,ns=ns,
                     model.terms=model.terms, model.formula=model.formula,
                     graph_mode=graph_mode,group=group,intercept=intercept,
                     exvar=exvar,
                     maxlag=maxlag,
                     lagmat=lagmat,
                     ylag=ylag,
                     lambda = NA, method='bayesglm',
                     alpha.glmnet=alpha.glmnet))
## End(Not run)

Simulation Engine for dynamic Vertex case.

Description

Simulation engine for dynamic networks with variable number of vertices. Implements exponential family based hierarchical model for vertices and the edges.

Usage

engineVertex(
  InputNetwork,
  numSim,
  maxLag,
  VertexStatsvec = rep(1, nvertexstats),
  VertexLag = rep(1, maxLag),
  VertexLagMatrix = matrix(1, maxLag, length(VertexStatsvec)),
  VertexModelGroup = NA,
  VertexAttLag = rep(1, maxLag),
  dayClassObserved = NA,
  dayClassFuture = NA,
  EdgeModelTerms,
  EdgeModelFormula,
  EdgeGroup = NA,
  EdgeIntercept = c("edges"),
  EdgeNetparam = NA,
  EdgeExvar = NA,
  EdgeLag = rep(1, maxLag),
  EdgeLagMatrix = matrix(1, maxLag, length(EdgeModelTerms)),
  regMethod = "bayesglm",
  paramout = TRUE
)

Arguments

Value

List with following elements: SimNetwork: Output Networks EdgeParameterMat: Matrix of edge parameter VertexParameterMat: Matrix of Vertex parameters.

Author(s)

Abhirup

Examples

## Not run: 
nvertexstats <- 9
maxLag = 3
VertexLag = rep(1, maxLag)
VertexLagMatrix <- matrix(0, maxLag, nvertexstats)
VertexLagMatrix[, c(4, 7)] <- 1
VertexLagMatrix[c(2,3),7] <- 0

getWeekend <- function(z){
    weekends <- c("Saturday", "Sunday")
    if(!network::is.network(z)){
        if(is.na(z)) return(NA)
    } else {
         zDay <- get.network.attribute(z, attrname = "day")
         out <- ifelse(zDay %in% weekends, 1, 0)
         return(out)   
    }
}

dayClass <- numeric(length(beach))
for(i in seq_along(dayClass)) {
    dayClass[i] <- getWeekend(beach[[i]])
}
dayClass <- na.omit(dayClass)
simResult <- suppressWarnings(engineVertex(InputNetwork = beach,
                          numSim = 5,
                          maxLag = 3,
                          VertexStatsvec = rep(1, nvertexstats),
                          VertexModelGroup = "regular",
                          VertexAttLag = rep(1, maxLag),
                          VertexLag = rep(1, maxLag),
                          VertexLagMatrix = VertexLagMatrix,
                          dayClassObserved = dayClass,
                          dayClassFuture = c(1, 0, 0, 0, 0),
                          EdgeModelTerms = NA,
                          EdgeModelFormula = NA,
                          EdgeGroup = NA,
                          EdgeIntercept = c("edges"),
                          EdgeNetparam = c("logSize"),
                          EdgeExvar = NA,
                          EdgeLag = c(0, 1, 0),
                          paramout = TRUE
                          ))
## End(Not run)

Simulation Engine for dynamic Vertex case without smoothing of estimated predictor matrices.

Description

Simulation engine for dynamic networks with variable number of vertices. Implements exponential family based hierarchical model for vertices and the edges. This does not implement smoothing for estimated predictor matrices.

Usage

engineVertexNS(
  InputNetwork,
  numSim,
  maxLag,
  VertexStatsvec = rep(1, nvertexstats),
  VertexLag = rep(1, maxLag),
  VertexLagMatrix = matrix(1, maxLag, length(VertexStatsvec)),
  VertexModelGroup = NA,
  VertexAttLag = rep(1, maxLag),
  dayClassObserved = NA,
  dayClassFuture = NA,
  EdgeModelTerms,
  EdgeModelFormula,
  EdgeGroup = NA,
  EdgeIntercept = c("edges"),
  EdgeNetparam = NA,
  EdgeExvar = NA,
  EdgeLag = rep(1, maxLag),
  EdgeLagMatrix = matrix(1, maxLag, length(EdgeModelTerms)),
  regMethod = "bayesglm",
  paramout = TRUE
)

Arguments

Value

List with following elements: SimNetwork: Output Networks
EdgeParameterMat: Matrix of edge parameter
VertexParameterMat: Matrix of Vertex parameters.

Examples

## Not run: 
nvertexstats <- 9
maxLag <- 3
VertexLag <- rep(1, maxLag)
VertexLagMatrix <- matrix(0, maxLag, nvertexstats)
VertexLagMatrix[, c(4, 7)] <- 1
VertexLagMatrix[c(2, 3), ] <- 1
simResult <- suppressWarnings(engineVertexNS(InputNetwork = beach,
                          numSim = 5,
                          maxLag = 3,
                          VertexStatsvec = rep(1, nvertexstats),
                          VertexModelGroup = "regular",
                          VertexAttLag = rep(1, maxLag),
                          VertexLag = rep(1, maxLag),
                          VertexLagMatrix = VertexLagMatrix,
                          EdgeModelTerms = NA,
                          EdgeModelFormula = NA,
                          EdgeGroup = NA,
                          EdgeIntercept = c("edges")
                          ))
## End(Not run)

expdeg

Description

Calculate the expectation of degree distribution of network

Usage

expdeg(x)

Arguments

Details

Given a network in adjacency matrix form, this calculates the expected degree statistic using igraph degree distribution function.

Value

scaler

Author(s)

Abhirup

Examples

expdeg(beach[[1]][, ])

ntriangles

Description

Calculate number of triangles of a network

Usage

ntriangles(x)

Arguments

Details

This function calculates the number of triangles in a network given an adjacency matrix. We use igraph for this.

Value

scaler, number of triangles

Author(s)

Abhirup

Examples

ntriangles(beach[[1]][, ])

Parameter estimation for static vertex case.

Description

Parameter estimation for the static vertex case.

Usage

paramEdge(
  input_network,
  model.terms,
  model.formula,
  graph_mode = "digraph",
  group,
  intercept = c("edges"),
  exvar = NA,
  maxlag = 3,
  lagmat = matrix(sample(c(0, 1), (maxlag + 1) * length(model.terms), replace = T),
    ncol = length(model.terms)),
  ylag = rep(1, maxlag),
  lambda = NA,
  method = "glmnet",
  alpha.glmnet = 1,
  paramout = TRUE
)

Arguments

Value

list with elements: coef: coefficients mplematfull: full matrix of change statistics mplemat: subset of matrix of change statistics

Author(s)

Abhirup

Examples

## Not run: 
input_network=rdNets[1:6]
model.terms=c("triadcensus.003", "triadcensus.012", "triadcensus.102", "triadcensus.021D", "gwesp");
model.formula = net~triadcensus(0:3)+gwesp(decay=0, fixed=FALSE, cutoff=30)-1;
graph_mode='digraph';
group='dnc';
alpha.glmnet=1
directed=TRUE;
method <- 'bayesglm'
maxlag <- 3
lambda=NA
intercept = c("edges")
cdim <- length(model.terms)
lagmat <- matrix(sample(c(0,1),(maxlag+1)*cdim,replace = TRUE),ncol = cdim)
ylag <- rep(1,maxlag)
exvar <- NA
out <- paramEdge(input_network,model.terms, model.formula,
                graph_mode='digraph',group,intercept = c("edges"),exvar=NA,
                maxlag = 3,
                lagmat = matrix(sample(c(0,1),(maxlag+1)*cdim,
                                       replace = TRUE),ncol = cdim),
                ylag = rep(1,maxlag),
                lambda = NA, method='bayesglm',
                alpha.glmnet=1)
## End(Not run)

Parameter estimation for Vertex dynamics

Description

Parameter estimation fro dynamic vertex case. The interface remaining almost identical to the static vertex one.

Usage

paramVertex(
  InputNetwork,
  VertexStatsvec = rep(1, nvertexstats),
  maxLag,
  VertexLag = rep(1, maxLag),
  VertexLagMatrix = matrix(1, maxLag, length(VertexStatsvec)),
  VertexModelGroup = NA,
  VertexAttLag = rep(1, maxLag),
  dayClass = NA,
  EdgeModelTerms,
  EdgeModelFormula,
  EdgeGroup,
  EdgeIntercept = c("edges"),
  EdgeNetparam = NA,
  EdgeExvar = NA,
  EdgeLag = rep(1, maxLag),
  EdgeLagMatrix = matrix(1, maxLag, length(EdgeModelTerms)),
  regMethod = "bayesglm",
  paramout = FALSE
)

Arguments

Details

The Vertex model parameter list is as follows (Freeman degree, In degree, Out degree, Eigen Centrality, Between centrality, Info centrality, Closeness centrality, log k cycles, log size). For more details about the definitions of the terms, please refer to the vertexstats.R file, which implements all of these. The definitions are in sna or igraph.

Value

list with following elements:
EdgeCoef: edge coefficients.
Edgemplematfull: MPLE matrix from edges.
Edgemplemat: Subsetted MPLE matrix.
VertexCoef: Coefficients from vertex.
Vstats: Vertex statistics matrix.
EdgePredictor0: Edge predictors with imputations with 0.
EdgePredictor1: Edge predictors with imputations with 1.
EdgePredictorNA: Edge predictors with imputations with NA.
EdgeFit: Edge model.
VertexStatsFull: Vertex statistics matrix, full.
VertexFit: Vertex model.

Author(s)

Abhirup

Examples

nvertexstats <- 9
maxLag = 3
VertexLag = rep(1, maxLag)
VertexLagMatrix <- matrix(0, maxLag, nvertexstats)
VertexLagMatrix[, c(4, 7)] <- 1
VertexLagMatrix[c(2,3),7] <- 0

getWeekend <- function(z){
    weekends <- c("Saturday", "Sunday")
    if(!network::is.network(z)){
        if(is.na(z)) return(NA)
    } else {
         zDay <- get.network.attribute(z, attrname = "day")
         out <- ifelse(zDay %in% weekends, 1, 0)
         return(out)   
    }
}

dayClass <- numeric(length(beach))
for(i in seq_along(dayClass)) {
    dayClass[i] <- getWeekend(beach[[i]])
}
dayClass <- na.omit(dayClass)


out <- paramVertex(InputNetwork = beach,
                   maxLag = 3,
                   VertexStatsvec = rep(1, nvertexstats),
                   VertexModelGroup = "regular",
                   VertexLag = rep(1, maxLag),
                   VertexLagMatrix = VertexLagMatrix,
                   dayClass = dayClass,
                   EdgeModelTerms = NA,
                   EdgeModelFormula = NA,
                   EdgeGroup = NA,
                   EdgeIntercept = c("edges"),
                   EdgeNetparam = c("logSize"),
                   EdgeExvar = NA,
                   EdgeLag = c(1, 1, 0),
                   paramout = TRUE)

Parameter estimation for Vertex model only for a list of dynamic networks.

Description

Parameter estimation for Vertex model only for a list of dynamic networks.

Usage

paramVertexOnly(
  InputNetwork,
  VertexStatsvec = rep(1, nvertexstats),
  maxLag,
  VertexLag = rep(1, maxLag),
  VertexLagMatrix = matrix(1, maxLag, length(VertexStatsvec)),
  dayClass = NA,
  regMethod = "bayesglm"
)

Arguments

Value

List of 3 elements:
VertexFit: Output from regEngine.
VertexStats: Subsetted vertex stats matrix.
VertexStatsFull: Full matrix of vertex stats.

Author(s)

Abhirup

Examples

nvertexstats <- 9
maxLag = 3
VertexLag = rep(1, maxLag)
VertexLagMatrix <- matrix(0, maxLag, nvertexstats)
VertexLagMatrix[, c(4, 7)] <- 1
VertexLagMatrix[c(2,3),7] <- 0
getWeekend <- function(z){
    weekends <- c("Saturday", "Sunday")
    if(!network::is.network(z)){
        if(is.na(z)) return(NA)
    } else {
         zDay <- get.network.attribute(z, attrname = "day")
         out <- ifelse(zDay %in% weekends, 1, 0)
         return(out)   
    }
}

## for(i in 1:31) print(getWeekend(beach[[i]]))
## generate a vector of network level exogenous variable
dayClass <- numeric(length(beach))
for(i in seq_along(dayClass)) {
    dayClass[i] <- getWeekend(beach[[i]])
}
out <- paramVertexOnly(InputNetwork = beach,
                       maxLag = 3,
                       VertexStatsvec = rep(1, nvertexstats),
                       VertexLag = rep(1, maxLag),
                       VertexLagMatrix = VertexLagMatrix,
                       dayClass = dayClass)

Parameter estimation for Vertex model only for a list of dynamic networks.

Description

Parameter estimation for Vertex model only for a list of dynamic networks.

Usage

paramVertexOnlyGroup(
  InputNetwork,
  VertexStatsvec = rep(1, nvertexstats),
  maxLag,
  VertexModelGroup = NA,
  VertexLag = rep(1, maxLag),
  VertexAttLag = rep(1, maxLag),
  VertexLagMatrix = matrix(1, maxLag, length(VertexStatsvec)),
  regMethod = "bayesglm"
)

Arguments

Value

List of 3 elements:
VertexFit: Output from regEngine.
VertexStats: Subsetted vertex stats matrix.
VertexStatsFull: Full matrix of vertex stats.

Author(s)

Abhirup

Examples

nvertexstats <- 9
InputNetwork <- beach
maxLag <- 3
VertexStatsvec <- rep(1, nvertexstats)
VertexLag <- rep(1, maxLag)
regMethod <- "bayesglm"
VertexModelGroup <- "regular"
VertexLagMatrix <- matrix(0, maxLag, nvertexstats)
VertexLagMatrix[, c(4, 7)] <- 1
VertexLagMatrix[c(2,3),7] <- 0
Vout1 <- paramVertexOnlyGroup(InputNetwork = beach,
                          maxLag = maxLag,
                          VertexStatsvec = VertexStatsvec,
                          VertexModelGroup = VertexModelGroup,
                          VertexLag = VertexLag,
                          VertexLagMatrix = VertexLagMatrix)
summary(Vout1$VertexFit$fit)

Blog citation network

Description

A data set of temporal inter and intra group blog citation network, with fixed number of vertex.

Usage

rdNets

Format

A list with 484 elements. Each element is a network of size 47 number of vertices.

Source

Butts, C. T. and B. R. Cross (2009). Change and external events in computer-mediated citation networks, English language weblogs and the 2004 u.s. electoral cycle. The Journal of Social Structure 10 (3), 1-29.

General purpose regression engine for the methods bayesglm, glm and glmnet

Description

General purpose regression engine for the methods bayesglm, glm and glmnet

Usage

regEngine(
  XYdata,
  method = "bayesglm",
  regIntercept = FALSE,
  lambda = NA,
  alpha = 1
)

Arguments

Value

list with elements: coef, se, lambda, fit (Coefficients, SE, lambda, if used, fit object.)

Author(s)

Abhirup

vdegree

Description

Calculates the degree of each vertices.

Usage

vdegree(x)

Arguments

Details

Given a network as adjacency matrix, calculate degree stats for each vertex.

Value

vector of length number of vertices.

Author(s)

Abhirup

Examples

vdegree(beach[[1]][, ])