examples-scripts/structural_reducibility.R

In manlius/muxViz: MuxViz - Visualization and Analysis of Multilayer Networks

library("muxViz")
library(igraph)

#Or build by specifying layers
Layers <- 10
Nodes <- 100
NodeTensor <- list()
g.list <- list()

g.list[[1]] <- barabasi.game(Nodes, m = 1, directed = F)

#generate additional layers with increasing % of overlapping edges, while preserving degree distribution
ov.fractions <- seq(0.1, 0.9, 0.1)
for (l in 2:10) {
  g.list[[l]] <-
    GetRewiredLayerWithOverlappingEdges(g.list[[1]], ov.fractions[l - 1], tol =
                                          1e-2)
}


#Define the layers by their adjacency matrices
for (l in 1:Layers) {
  NodeTensor[[l]] <- get.adjacency(g.list[[l]])
}

#Define how layers are interconnected
LayerTensor <- Matrix::Matrix(1, Layers, Layers) - muxViz:::speye(Layers)

#Build the multilayer adjacency tensor
M <-
  BuildSupraAdjacencyMatrixFromEdgeColoredMatrices(NodeTensor, LayerTensor, Layers, Nodes)

#use Ward linkage for hierarchical clustering and Categorical network of layers
Method <- "ward.D2"
Type <- "Categorical"

struct.red <- GetMultilayerReducibility(M, Layers, Nodes, Method, Type)

png(
  "mux_structural_reducibility_hclust.png",
  width = 1024,
  height = 1024,
  res = 120
)
gplots::heatmap.2(as.matrix(struct.red$JSD), trace = "none")
dev.off()

df.quality <-
  data.frame(step = 0:(length(struct.red$relgQualityFunction) - 1),
             q = struct.red$relgQualityFunction)
png(
  "mux_structural_reducibility_quality.png",
  width = 1024,
  height = 1024 * 0.5,
  res = 120
)
p <- ggplot(df.quality, aes(step, q)) + theme_bw() +
  geom_line(color = "steelblue") + geom_point(color = "steelblue") +
  xlab("Merging Step, m") + ylab("Quality function, q(m)")
print(p)
dev.off()