During clustering and appropriate visualization of developer networks,
it is necessary to find the most important node in a cluster (the one
with the highest rank, given a list of ranks). But, the current
implementation for identification of the most important node per
cluster actually only works for non-duplicate ranks.
If there are two nodes with the same rank which are in different
clusters and the most important node for their respective cluster, it
may be the case that the wrong one is picked. The new statements ensure
that exactly the one most important node from the current community is
selected, and not any node that has the same rank and, accidentally,
comes first in the node order.
Plus, remove trailing whitespace.
[CH: Adjust commit message.]
Signed-off-by: Thomas Bock <bockthom@xxxxxxxxxxxxxxxxx>
Reviewed-by: Claus Hunsen <hunsen@xxxxxxxxxxxxxxxxx>
---
codeface/R/cluster/network_visualization.r | 20 ++++++++++----------
1 file changed, 10 insertions(+), 10 deletions(-)
diff --git a/codeface/R/cluster/network_visualization.r
b/codeface/R/cluster/network_visualization.r
index 1fae8b6..d1c41b5 100644
--- a/codeface/R/cluster/network_visualization.r
+++ b/codeface/R/cluster/network_visualization.r
@@ -43,8 +43,8 @@ important.community.nodes <- function(comm, rank) {
community.ids <- sort(unique(comm$membership))
important.nodes <- sapply(community.ids,
function(comm.id) {
- which(rank ==
- max(rank[which(comm$membership == comm.id)]))[1]
+ node.ids.in.comm.id <- which(comm$membership == comm.id)
+ node.ids.in.comm.id[which.max(rank[node.ids.in.comm.id])]
})
return (important.nodes)
}
@@ -167,8 +167,8 @@ computeVertCommFrac <- function (graph, comm) {
## percentages are captured.
verts.frac <- list()
for (i in V(graph)) {
- ## Get neighbors of vertex i, mode=all will return both in and out
- ## directions, multiple edges are listed multiple times however we want
+ ## Get neighbors of vertex i, mode=all will return both in and out
+ ## directions, multiple edges are listed multiple times however we want
## the unique vertex index
total <- 0
comm.frac <- list()
@@ -183,7 +183,7 @@ computeVertCommFrac <- function (graph, comm) {
if(length(comm.frac[[key]]) == 0) {
comm.frac[[key]] <- 0
}
- ## get edge weight and sum directions (in weight + out weight)
+ ## get edge weight and sum directions (in weight + out weight)
edge.weight <- sum(E(graph)[i %--% vert.neigh[j]]$weight)
comm.frac[[key]] <- comm.frac[[key]] + edge.weight
}
@@ -304,7 +304,7 @@ compute.subgraph.list <- function (g, comm){
community.id <- sort(unique(comm$membership))
subgraphL <- list()
subgraphL <- lapply(community.id,
- function(x) {
+ function(x) {
return(list(graph=subGraph(as.character(which(comm$membership==x)), g)))
})
return(subgraphL)
@@ -316,14 +316,14 @@ format.color.weight <- function(colorL, weightL) {
## for the pie chart style node. The graphviz library demands the following
## color format WC:WC:WC:WC where WC = color;weight
color.weight.list <- mapply(
- function(colors, weights) paste(paste(colors, as.character(weights),
sep=";"),
+ function(colors, weights) paste(paste(colors, as.character(weights),
sep=";"),
collapse=":"),
colorL, weightL, SIMPLIFY=FALSE)
-
+
## Add ":" to the end of every string, for some reason if the weight is 1
## and there is no ":" after, graphviz will not fill the node with color
color.weight.vec <- paste(unlist(color.weight.list), ":", sep="")
-
+
return(color.weight.vec)
}
@@ -425,7 +425,7 @@ save.graph.graphviz <- function(con, pid, range.id,
cluster.method, filename,
To <- as.character(get.edgelist(g.min.simp)[,2])
edge.df <- data.frame(From=From, To=To)
weights <- E(g.min.simp)$weight
- g.NEL <- ftM2graphNEL(as.matrix(edge.df), W=weights, V=V(g.min.simp)$name,
+ g.NEL <- ftM2graphNEL(as.matrix(edge.df), W=weights, V=V(g.min.simp)$name,
edgemode="directed")
subgraph.list <- compute.subgraph.list(g.NEL, comm)
g.viz <- agopen(g.NEL, "pieGraph", subGList=subgraph.list)
--
2.8.0
