Build and analyze network of CRAN packages

2016-08-12 16:57:01

Package DESCRIPTION files allows for specifying several types of inter-package relations. These include fields like Depends, Suggests, Enhances etc. This package allows for recovering graph structure based on these relations. Network representation of R repositories enables the user to explore the interconnected space of available R functionality while the developers or repository maintainers can scan package forward and reverse dependencies etc.

The main function pkgnet creates an igraph object of packages available in specified repository containing all other fields from available.packages as vertex and edge attributes.

Examples

Network based on current snapshot of CRAN

This will fetch data on packages available on CRAN at this very moment and transform it to an igraph object:

g <- pkgnet("cran")
summary(g)

## IGRAPH DN-- 9186 47048 -- 
## + attr: name (v/c), Version (v/c), Priority (v/c), License (v/c),
## | License_is_FOSS (v/c), License_restricts_use (v/c), OS_type
## | (v/c), Archs (v/c), MD5sum (v/c), NeedsCompilation (v/c), File
## | (v/c), Repository (v/c), type (e/c)

Thus today (2016-08-12) we have 9186 packages in total. This however includes also packages that are not available on CRAN, but are mentioned by other packages in their DESCRIPTION files. This can be checked using the Repository field which will be equal to NA for packages not available on CRAN:

sum( is.na(V(g)$Repository) )

## [1] 248

Graph of packages related to network analysis

Let extract a subgraph of packages depending (Depends, Imports, or Suggests) on one of the packages providing facilities to store and process network data ("graph", "igraph", "network", "networkDynamic").

Get the neighborhoods and make the graph:

# Seed packages
seeds <- c("igraph", "network", "graph")

# Only dependence arcs
depnet <- delete.edges(g, E(g)[ !(type %in% c("Depends", "Imports", "Suggests")) ])

# list of package names
pkg_names <- ego(depnet, 1, V(depnet)[ name %in% seeds ], mode="in" )
sg <- simplify( induced_subgraph(depnet, unique(unlist(pkg_names))) )
summary(sg)

## IGRAPH DN-- 344 563 -- 
## + attr: name (v/c), Version (v/c), Priority (v/c), License (v/c),
## | License_is_FOSS (v/c), License_restricts_use (v/c), OS_type
## | (v/c), Archs (v/c), MD5sum (v/c), NeedsCompilation (v/c), File
## | (v/c), Repository (v/c)

Plot it such that:

• Arcs go "upstream"
• Nodes for seed packages are larger
• Colors show which of the "seed" packages a given package depends on

set.seed(2992)

# Coordinates
xy <- layout.fruchterman.reingold(sg)

# Colors
d <- distances(sg, to=V(sg)[ name %in% seeds], mode="out")
x <- apply(d, 1, function(x) paste(sort(colnames(d)[x == min(x)]), collapse="+"))
pal <- RColorBrewer::brewer.pal(length(unique(x)), "Set1")
cols <- pal[ match(x, sort(unique(x)))]

# Plot
plot(sg, 
     vertex.size=ifelse( V(sg)$name %in% seeds, 10, 5),
     vertex.color=cols,
     layout=xy, 
     vertex.label.color="black",
     # vertex.frame.color = par("bg"),
     edge.arrow.size=.5, 
     vertex.label=NA,
     margin=0
     )
legend("bottomright",
       title="Dependencies",
       sort(unique(x)),
       pch=21,
       pt.bg=pal,
       col="black",
       bty="n"
       )

Installation

devtools::install_github("mbojan/pkgnet")