Save Network: save the same as in NetworkX

[ajdapretnar]

Network version

1.9.0

Orange version

3.37.0

Expected behavior

Save Network saves the data (attributes) in the same way as etworkX would. See below.

Actual behavior

The data is saved as such. Orange can't read what it saves (attributes not loaded properly). NetworkX can't read what Orange saves. Also, Orange can't read what NetworkX saves.

Steps to reproduce the behavior

import networkx as nx

Create a graph

G = nx.Graph()

# Add nodes with attributes
G.add_node(1, name='Alice', age="30")
G.add_node(2, name='Bob', age="25")
G.add_node(3, name='Charlie', age="35")
G.add_node(4, name='David', age="40")

# Add edges
G.add_edge(1, 2)
G.add_edge(1, 3)
G.add_edge(2, 3)
G.add_edge(3, 4)

nx.write_pajek(G, "simple_network.net")

Additional info (worksheets, data, screenshots, ...)

While pajek doesn't have a documentation on its format (from what I know), it would be nice to have the format transferable between NetworkX and Orange. Or at the very least for Orange to properly read the data it saves.

[janezd]

This shouldn't be too difficult to implement. A few observations, though.

• Orange's reader currently ignores everything after coordinates, label, x, y, *_ = parts + [None, None]. This is good; we didn't add any custom data (in the meanwhile Network File and Explorer: Load data for edges; show colored and labelled edges #249 allows giving labels to edges).
• Networkx seems to allow arbitrary key-value pairs after nodes. In Orange, this would be represented as a Table, hence every key would become a variable, and variable types would be deducted from values, I guess, using the same rules as one-line tab files.
• Besides key-value pairs, networkx adds a shape (ellipsis). This is probably something like a reserved word. Do you know what are the others?
• Can you check why Networkx can't read files that are created in Orange? I don't think we're adding anything fancy.

[ajdapretnar]

Can you check why Networkx can't read files that are created in Orange? I don't think we're adding anything fancy.

When I said it "can't read files" I was being dramatic. I meant it doesn't handle attributes properly. It takes everything as a node label, i.e. '[544 | G1] {security, G1}'. Ideally, the only node label would be "security", with "544" as node attribute Count (which was present in the original data). And, of course, "G1" would be an attribute Group, so that the user could navigate between attributes properly. Now it is a single item.

[janezd]

Note to self.

I've been looking into this, again. Pajek format is one big mess (which is certainly true for many of formats I defined or participated in defining.) For instance, while Vlado writes that node id and label are followed by two coordinates (http://vlado.fmf.uni-lj.si/pub/networks/pajek/doc/draweps.htm), Andrej says there are three (http://mrvar.fdv.uni-lj.si/pajek/DrawEPS.htm). Networkx uses the former, but then it also (ab?)uses the shape attributes to add arbitrary data in form of key-value pairs split by spaces.