计算非常大的网络的图表效率指标

问题描述 投票:0回答:1

What are strategies for calculating the nodal and global efficiency of very large graphs in R?

我试图用igraph计算一个非常大的brainGraph::efficiency(my_graph, type = "global")的全球效率。

library(igraph); library(brainGraph)  
g <- readRDS("path_to_my_graph.rds")  

> ecount(g); vcount(g) # count of edges and vertices
[1] 715758
[1] 290190

它每次都可靠地崩溃R.全球效率是所有节点效率的平均值,所以我试图以这种方式计算它并没有成功。我的图形每条边上的权重都是1,所以我省略了权重,但R仍然崩溃。

# all of these cause R to crash
efficiency(g, type = "global")
efficiency(g, type = "nodal")
efficiency(g, type = "global", weights = NA)
efficiency(g, type = "nodal",  weights = NA)

对于那些想要测试数据的人来说,我的图表(~37MB)可用here on GoogleDrive as an .rds file

r igraph network-efficiency
1个回答
0
投票

R崩溃,因为brainGraph::efficiency()试图计算一个巨大而密集的距离矩阵,它压倒了我机器的内存(32 GB)。但我发现了一个解决方案,可以将操作分块并行运行。

全局效率是图中所有节点效率的平均值。顶点i的节点效率是:

enter image description here

我们可以顺序计算图中每个顶点的节点效率,将距离矩阵计算分成较小的可管理位。因为每个顶点的效率是独立的,所以我们可以并行化操作,因此不需要永远。

library(igraph)  
library(doParallel)

# nodal efficiency function
get_eff <- function(i){return((1/(length(V(g)) - 1))*sum(1/distances(g, V(g)[i])[-i]))}

no_cores <- detectCores() - 1 
cl       <- makeCluster(no_cores)
registerDoParallel(cl)  

result <- foreach(i = seq_along(V(g)), .combine = c, .packages = "igraph") %dopar% get_eff(i)

stopCluster(cl)
rm(cl)

global_eff <- mean(result)

此外,我们可以绘制节点效率的分布以及全局(平均)效率,这使我们更好地了解网络。

library(ggplot2)
data.frame(x = result) %>% 
  ggplot(aes(x)) + 
  geom_histogram() + 
  geom_vline(xintercept = mean(result), col = "red") # global efficiency
  theme_minimal() +
  labs(title = "Nodal efficiences", x = "", y = "Count")

enter image description here

© www.soinside.com 2019 - 2024. All rights reserved.