Skip to main content

Graph Formats

Many file formats exist for representing graphs, including GraphML, GEXF, GML,
LEDA, Pajek, SparseGraph6, etc ... If you want to import these into a
Zef graph, this is the place to learn how.

Instead of reinventing the wheel, we make use of NetworkX to first parse the
file into memory.

In this example, we'll use a example GraphML file of TODO: sample.graphml

Load into NetworkX​

import networkx as nx  
  
ng = nx.read_graphml("sample.graphml")

Import into a Zef graph​

You can import the NetworkX graph as entities/relations on any Zef graph. Here
we start with a blank one: 

from zef import *  
from zef.ops import *  
from zef.experimental.importers.general import inject_networkx_into_zef  
  
zg = Graph()  
inject_networkx_into_zef(ng, zg)

The return value of the inject_networkx_into_zef function is a tuple of two
dictionaries, which map the ids in the NetworkX graph to the ZefRefs. 

zg = Graph()  
n_map,e_map = inject_networkx_into_zef(ng, zg)  
  
n_map["n0"] | Out[RT.Color] | value | run[print]  
green

Import options​

Translation options​

The import requires creating entities and relations and these must have types
in the Zef graph (ET.x and RT.y). However, in the imported data these are
type-less nodes and edges. You can control how the importer guesses the
translation of these using the keyword options:

  • ET_translation
  • RT_translation
  • fieldname_translation
  • aet_translation

The default translation tries to guess the ET or RT from the ID. But there are
plenty of times where this is not right. For example, this TODO: sample data
has only numeric IDs, which will convert to ET.Node
in the default translation. You might want to instead import this as: 

inject_networkx_into_zef(ng, zg,  
                         ET_translation=always["Person"],  
                         RT_translation=always["Knows"],  
                         ignore_excess_ET_RT_types=True)

This makes all entities ET.Person and all relations between entities
RT.Knows. The relation types for fields remain the same, determined from the
key, e.g. RT.Name and RT.Age.

Store ID option​

The keyword argument store_id is set to RT.ID by default, meaning that node
ids from the NetworkX graph are added to the created Zef entities by a RT.ID
relation. If you want to disable this you can set store_id=None.

Exporting back to the same format​

This How-To is about imports only. Zef graphs in general can be meta-graphs and
so might not be directly representable in any of the formats. However, one
option we have is to use the ZefDoc - NetworkX Interface to view and then export your Zef
graph to a file. For example: 

  
from zef.experimental.networkx import ProxyGraph  
import networkx as nx  
  
ng = ProxyGraph(now(zg), include_type_as_field=False)  
ng_native = ng.to_native(include_fields=True)  
nx.write_graphml(ng_native, "output.graphml")

Note that this will not reproduce the imported file exactly, unless some
significant customization is applied to the ProxyGraph functions.