Re: [igraph] R: vertex attributes in bipartite graphs for only one vertex type

[Dan Suthers]

Hi all,

The blind trust in V(g)$type is something that can trip you up. I have a 
large network (constructed outside of igraph) in which I had used a 
character vertex attribute 'type' to encode the entity type of nodes in 
the network (representing actors, chat rooms, files, discussion forums 
...). Igraph summary prints it as "bipartite" because it sees a 'type' 
attribute: does not even check that this attribute is logical.

> summary(ti)
IGRAPH e7dc05c D-WB 40490 229072 --
+ attr: label (v/c), type (v/c), data1 (v/c), data2 (v/c), data3
| (v/c), id (v/c), Edge Label (e/c), weight (e/n), Edge Id (e/c)
> head(V(ti)$type)
[1] "ACTOR" "ACTOR" "ROOM"  "ACTOR" "ACTOR" "ROOM"

No problem: we can supply a 'types' vector when calling 
bipartite_projection. I used it with V(g)$type == "ACTOR" to separate 
the human actors from the media types. But here is the 'gotcha': 
bipartite_projection has default remove.type = TRUE, so my data on the 
different media types is removed from the media object projection! One 
can set remove.type = FALSE, or to avoid future problems rename 'type' 
to something like 'entity'.

I deliberately left this in a homework problem I gave students, to give 
them practice with identifying and solving issues, as data cleaning will 
always be with us.

Your other issue is also illustrated with this network. Actors, Rooms, 
Discussions, and Files all have different attributes, but if we put an 
attribute such as 'gender' in your example (mine were different) on 
ACTOR, then the other nodes all get this attribute too. Similarly all 
nodes would have a MIME type attribute even though they only apply to 
files. This is because the underlying representation is tabular. The 
same is true of Gephi, which I usually use to inspect results of 
analyses. As you can see in the summary above, we avoided the 
proliferation of columns in Gephi Data Laboratory by re-using generic 
'data1', 'data2' and 'data3' columns, the contents of which depend on 
'type'. This was OK for visual inspection in Gephi data laboratory, but 
is not optimal.

Dan Suthers

On 11/25/19 8:04 PM, Michael Gastner wrote:
> Dear Tamas,
>
> Many thanks for your reply! Also thank you for explaining that igraph
> does not check whether the topology is really bipartite. I had been
> wondering about that too.
>
> Michael
>
> > > Is there some way to create a vertex attribute that is only attached
> > > to one vertex type in a bipartite graph (in this example the vertices
> > > for which V(g)$type == FALSE)?
> > Unfortunately there's no way to do that; this is because bipartite graphs
> > in igraph are simply "normal" graphs with a "type" vertex attribute and an
> > additional property that edges never go between vertices of the same type.
> > (But this is not even enforced; you can create such an edge if you really
> > want to). So, for the time being you need to settle with having NA as the
> > "gender" for the company vertices. Try to think about NA as "not
> > applicable", which actually sort-of-makes sense for companies :)
> _______________________________________________
> igraph-help mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/igraph-help

--
Dan Suthers

Professor and Graduate Program Chair
Dept. of Information and Computer Sciences
University of Hawaii at Manoa
1680 East West Road, POST 309, Honolulu, HI 96822
(808) 956-3890 office
Personal: http://www2.hawaii.edu/~suthers/
Lab: http://lilt.ics.hawaii.edu/
Department: http://www.ics.hawaii.edu/