igraph compute metrics for each node and its network

Question

I am curious how to compute some metrics for each node.

For each node compute percentage of fraudulent connections for

direct node (directed)
direct node (undirected)
the friendship network from the node (directed)
the friendship network from the node (undirected)

in total and per relationship type.

Getting started with igraph I am not sure how to move forward to writing own graph processing functions (i.e. not only applying degree, pagerank, ...). Looking forward to some suggestions to solve this task with only one pass over the graph.

Minimal sample is here

library(igraph)
id = c("a", "b", "c", "d", "e", "f", "g") 
name = c("Alice", "Bob", "Charlie", "David", "Esther", "Fanny", "Gaby") 
fraud = c(TRUE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE) 
verticeData <- data.frame(id, name, fraud) 
verticeData

src <- c("a", "b", "c", "f", "e", "e", "d", "a")
dst <- c("b", "c", "b", "c", "f", "d", "a", "e")
relationship <-c("A", "B", "B", "B", "B", "A", "A", "A")
edgeData <- data.frame(src, dst, relationship)
edgeData
g <- graph_from_data_frame(edgeData, directed = TRUE, vertices = verticeData)
plot(g, vertex.color=V(g)$fraud)
# TODO compute metrics

I do not have privileges to move, so will do manually based on comment from https://stats.stackexchange.com/questions/256859/igraph-compute-metrics-for-each-node-and-its-network

paqmo paqmo · Accepted Answer · 2017-01-18T16:58:06

The gapply function from the sna package gives a lot of flexibility to calculate various ego network statistics. It functions more or less like the apply family of functions, but specifically loops over network neighborhoods. The intergraph package makes it easy to convert between igraph and sna.

library(sna)
net<-intergraph::asNetwork(g)
c <- c(1,2)
funcs <- c(sum,mean)
for (i in funcs){
  for (j in list(1,2,c)){
    print(gapply(net,j,net %v% "fraud",i)) 
  }
}

gapply in not super straight forward to use. The second argument ("MARGIN") indicates either row-wise (outgoing ties), column-wise (incoming ties), or both (i.e., undirected). The third argument is a vector of statistics to calculate, and the fourth argument is the function you want to use. As you can, there is a lot of flexibility in the third and fourth arguments.

> gapply(net,c(1,2),net %v% "fraud",sum)
[1] 0 1 0 1 1 0 0
> gapply(net,c(1),net %v% "fraud",sum)
  Alice     Bob Charlie   David  Esther   Fanny    Gaby 
      0       0       0       1       0       0       0 
> gapply(net,c(2),net %v% "fraud",sum)
  Alice     Bob Charlie   David  Esther   Fanny    Gaby 
      0       1       0       0       1       0       0

igraph compute metrics for each node and its network

1 Answers