Serialize and unserialize in R. How to coerce similar objects?

Usage
1

I tried the following:

library(yaml)
library(igraph)

g1 <- make_star(7)
y1 <- as.yaml(g1)
g2 <- yaml.load(y1)

class(g1)                   # = igraph
class(g2)                   # = list

identical(str(g1), str(g2))	# true

g1 and g2 are structurally identical. But g1 has the class igraph. And g2 the class list. Is there an obvious way to convert g2 to igraph class.

2

No, there isn’t, because not all data is saved during the as.yaml() step:

> g1 <- make_star(7)
> unclass(g1)[[10]]
<environment: 0x10a0252e0>
> g2 <- yaml.load(as.yaml(g1))
> g2[[10]]
Error in g2[[10]] : subscript out of bounds

The part that is lost is a hidden R environment that stores all sorts of stuff about the graph that igraph itself needs:

> ls(unclass(g1)[[10]], all.names=T)
[1] ".__igraph_version__." "myid"

I guess you’ll need saveRDS() and readRDS() if you want to store an igraph object into a file in a way that preserves all R-specific stuff.

1 Like
3

Then, is it possible to "un.igraph" an igraph object and retrieve the underlying list structure?
Omitting the platform dependent data, but enough to reconstruct the graph.

As an example for graph g1: g2 <- as.igraph(un.igraph(g1))

If this is possible, then making a nice print of an igrapgh object is equivalent to printing a (nested) list neatly: prettyprint(un.igraph(g1)).

What I’m actually asking is a cross-platform definition of an igraph object.

4

As we said before, saveRDS / readRDS provide a cross-platform way to losslessly serialize an R/igraph object.

You focus on “serializing igraph objects”, but since graph attributes can be arbitrary R objects, the only way to do this is to use a serialization that works for any R object. That naturally leads to save / load / saveRDS / readRDS.

5

There are also functions like dput and dump, but as the dput documentation states,

Deparsing an object is difficult, and not always possible. With the default control , dput() attempts to deparse in a way that is readable, but for more complex or unusual objects (see dump ), not likely to be parsed as identical to the original. Use control = "all" for the most complete deparsing;

None of this is igraph specific though. It all applies to any R object.

While dput is nice for getting a fairly accurate picture of what is in an R object, the documentation states that it cannot handle all objects and is not suitable for data transfer:

This is not a good way to transfer objects between R sessions. dump is better, but the functions save and saveRDS are designed to be used for transporting R data, and will work with R objects that dput does not handle correctly as well as being much faster.

So, straight from the horse’s mouth (the R documentation), there’s your answer: use save or saveRDS.

6

Thanks for the very helpful comments.

What I’m trying to achieve is to distribute an object, but in a concise, human-readable form.
With minimal dependencies on system and environment. So that others can experiment with the result. And in such a way that it can be published in limited space.

Building on the received input I arrived at:

library(igraph)
g1 <- make_star(7)
dput(g1) # transfer output to the code example

# code example
library(igraph)
s1 = r"---(
structure(list(7, TRUE,
 c(1, 2, 3, 4, 5, 6),
 c(0, 0, 0, 0, 0, 0),
 c(0, 1, 2, 3, 4, 5),
 c(0, 1, 2, 3, 4, 5),
 c(0, 0, 1, 2, 3, 4, 5, 6),
 c(0, 6, 6, 6, 6, 6, 6, 6),
 list(c(1, 0, 1),
 list(name = "In-star",  mode = "in", center = 1),
 list(),
 list()) ),
 class = "igraph") 
)---"
g2 <- dget(rawConnection(charToRaw(s1)), keep.source=FALSE) 
g2 <- upgrade_graph(g2)
stopifnot(identical_graphs(g1, g2))

Dget doesn’t like <envionment> so I got rid of that.

7

Using a raw vector can be avoided, making it simpler.

library(igraph)
g1 <- make_star(7)
dput(g1) # transfer output to the code example

# code example
library(igraph)
s1 = deparse(
structure(list(7, TRUE,
 c(1, 2, 3, 4, 5, 6),
 c(0, 0, 0, 0, 0, 0),
 c(0, 1, 2, 3, 4, 5),
 c(0, 1, 2, 3, 4, 5),
 c(0, 0, 1, 2, 3, 4, 5, 6),
 c(0, 6, 6, 6, 6, 6, 6, 6),
 list(c(1, 0, 1),
 list(name = "In-star",  mode = "in", center = 1),
 list(),
 list()) ),
 class = "igraph") 
)
g2 <- dget(textConnection(s1), keep.source=FALSE) 
g2 <- upgrade_graph(g2)
stopifnot(identical_graphs(g1, g2))
8

Triggered by Tamas’s example, I think it’s better to avoid the dput of the igraph object altogether by applying an unclass operation first.

library(igraph)
g1 <- graph_from_literal(1 -+2)		# non-empty igraph object
ug <- unclass(g1)					# remove class igraph
ug[[10]] <- NULL					# remove environment
dput(ug, stdout())				    # output to console

us <- list(2, TRUE, 0, 1, 0, 0, c(0, 1, 1), c(0, 0, 1), list(c(1, 0, 1), structure(list(), .Names = character(0)), list(name = c("1", "2")), structure(list(), .Names = character(0))))

g2 <- structure(us, class="igraph") # add class igraph
g3 <- upgrade_graph(g2)				# and upgrade to current environment