graph_from_data_frame does not simplify graph

[SamGG]

Hi,
Thanks for maintaining the important igraph package giving a uniform access to many algorithms.
I think that the graph_from_data_frame function is not explicit concerning the directed argument. When entering a not directed graph with duplicated links, one might expect that the graph will be simplified implicitly. It is not the case, and I think this should be clearly stated in the documentation.
I don't know how an undirected and unsimplified graph is handled by community detection algorithms.
Best.

library(igraph)
#> 
#> Attaching package: 'igraph'
#> The following objects are masked from 'package:stats':
#> 
#>     decompose, spectrum
#> The following object is masked from 'package:base':
#> 
#>     union
df = data.frame(from=c(1,2,3,4,1), to=c(2,1,3,5,2), weight=c(1,1,1,1,3))

g <- graph_from_data_frame(df)
g
#> IGRAPH 677ae70 DNW- 5 5 -- 
#> + attr: name (v/c), weight (e/n)
#> + edges from 677ae70 (vertex names):
#> [1] 1->2 2->1 3->3 4->5 1->2
vertex.attributes(g)
#> $name
#> [1] "1" "2" "3" "4" "5"

g <- graph_from_data_frame(df, directed = FALSE)
g
#> IGRAPH 677bf7b UNW- 5 5 -- 
#> + attr: name (v/c), weight (e/n)
#> + edges from 677bf7b (vertex names):
#> [1] 1--2 1--2 3--3 4--5 1--2
edge.attributes(g)
#> $weight
#> [1] 1 1 1 1 3

h <- simplify(g, remove.multiple = FALSE, edge.attr.comb = "sum")
h
#> IGRAPH 677ce17 UNW- 5 4 -- 
#> + attr: name (v/c), weight (e/n)
#> + edges from 677ce17 (vertex names):
#> [1] 1--2 1--2 4--5 1--2
edge.attributes(h)
#> $weight
#> [1] 1 1 1 3

h <- simplify(g, edge.attr.comb = "sum")
h
#> IGRAPH 677d7d4 UNW- 5 2 -- 
#> + attr: name (v/c), weight (e/n)
#> + edges from 677d7d4 (vertex names):
#> [1] 1--2 4--5
edge.attributes(h)
#> $weight
#> [1] 5 1

^{Created on 2020-05-23 by the reprex package (v0.3.0)}

[mbojan]

I don't think graph_from_data_frame() should simplify the graph. There are many settings in which you do not want that and want to make the graph with multiple edges as they are in the edgelist. It is IMHO the right approach in your example that you

graph_from_data_frame(...) %>%
  simplify(...)

if you need to analyze a simple graph instead of a multi-graph.

[gaborcsardi]

It is clear that we cannot change this function now:

• it would break existing code,
• as @mbojan argues, sometimes you do need those multiple edges, and
• it is not clear what to do with the edge attributes when simplifying the graph.

@SamGG if you think this is surprising behavior, we can add a note to the manual page. Would you like to submit a pull request?

[SamGG]

Thanks for your feedback. I didn't intend to change the behavior of this function, but just to make the documentation more clear about that. I will do a PR.

[ntamas]

The documentation of graph_from_data_frame now includes this sentence:

All edges in the data frame are included in the graph, which may include multiple parallel edges and loops.

I don't know when this was added, but I consider the issue as fixed.