GraphML does not support type <class 'type'> as data values.

[scls19fr]

Hello,

I tried to export a graph from ArangoDB to graphml thanks to NetworkX and networkx-adapter

I did

from dotenv import load_dotenv
from arango import ArangoClient
from adbnx_adapter import ADBNX_Adapter
load_dotenv()
client = ArangoClient(hosts="http://localhost:8529")
db = client.db("My-DB", username="root", password=os.getenv("ARANGO_ROOT_PASSWORD"))
adbnx_adapter = ADBNX_Adapter(db)
nx_g = adbnx_adapter.arangodb_graph_to_networkx("My-Graph")
nx.write_graphml_lxml(nx_g, "my.graphml")

but it raised

GraphML does not support type <class 'type'> as data values.

I think my nodes have attributes named type which doesn't work well with such an export.

So I tried

def convert_data_types(data):
    """Converts unsupported data types to strings."""
    if isinstance(data, type):
        return str(data)
    else:
        return data

#print("nodes")
for node, attr_dict in nx_g.nodes(data=True):
    nx_g.nodes[node]["attributes"] = {key: convert_data_types(value) for key, value in attr_dict.items()}

#print("edges")
for edge in nx_g.edges(data=True):
    source, target, attr_dict = edge
    d = {key: convert_data_types(value) for key, value in attr_dict.items()}
    nx_g.edges[source][target] = d

nx.write_graphml_lxml(nx_g, "gh.graphml")

but it raises

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[67], line 16
     14     source, target, attr_dict = edge
     15     d = {key: convert_data_types(value) for key, value in attr_dict.items()}
---> 16     nx_g.edges[source][target] = d
     18 nx.write_graphml_lxml(nx_g, "gh.graphml")

File ~\anaconda3\Lib\site-packages\networkx\classes\reportviews.py:1352, in OutMultiEdgeView.__getitem__(self, e)
   1347 if isinstance(e, slice):
   1348     raise nx.NetworkXError(
   1349         f"{type(self).__name__} does not support slicing, "
   1350         f"try list(G.edges)[{e.start}:{e.stop}:{e.step}]"
   1351     )
-> 1352 u, v, k = e
   1353 return self._adjdict[u][v][k]

ValueError: too many values to unpack (expected 3)

What is the most efficient way to fix that?

Kind regards

PS : also posting at arangoml/networkx-adapter#96

[Schefflera-Arboricola]

I think your convert_data_types should be used on keys, not values, and this:

nx_g.nodes[node]["attributes"] = {key: convert_data_types(value) for key, value in attr_dict.items()}

will create a node attribute named attributes with this dictionary as the value.
Also, I don't think one can change an attribute's name in networkx(ref. https://stackoverflow.com/questions/69423243/change-an-attribute-name-in-networkx). You would have to create a new attribute and delete the old one.

Thank you :)

[scls19fr]

Well... you are right. I should apply function on keys... not on values. Thanks!

but when I'm doing

def convert_key(data):
    return "node-type" if data == "type" else data

for node, attr_dict in nx_g.nodes(data=True):
    nx_g.nodes[node]["attributes"] = {convert_key(key): value for key, value in attr_dict.items()}

for edge in nx_g.edges(data=True):
    source, target, attr_dict = edge
    if "type" in attr_dict:
        attr_dict["node-type"] = attr_dict.pop("type")

nx.write_graphml_lxml(nx_g, "gh.graphml")

it still raises

  TypeError: GraphML does not support type <class 'type'> as data values.

as if I didn't changed anything

I unsure if I shouldn't use nx.set_edge_attributes (and how to use it with my MultiDiGraph)

[scls19fr]

The type attribute is in fact only present in nodes... not edges. So I did

for node, attr_dict in nx_g.nodes(data=True):
    new_attr_dict = {convert_key(key): value for key, value in attr_dict.items()}
    nx.set_node_attributes(nx_g, new_attr_dict, name=node)

but when I print nodes in an other for loop ... they are not modified ie attr_dict still have type key.

That's very difficult to me to understand what is going on.

[scls19fr]

I think it will be easier to post you a minimal working example with sample data.

I have been able to output at .dot file like so gh.dot

digraph "VS-GH-Graph" {
	graph [name="VS-GH-Graph"];
	"Users/user0"	[_id="Users/user0",
		_key=user0,
		_rev="_hzGd4vu---",
		avatar_url="",
		bio="My bio",
		blog="",
		collaborators=11,
		company="",
		created_at="2009-07-27T18:53:24Z",
		disk_usage=678928,
		email=None,
		events_url="https://api.github.com/users/user0/events{/privacy}",
		followers=79,
		followers_url="https://api.github.com/users/user0/followers",
		following=248,
		following_url="https://api.github.com/users/user0/following{/other_user}",
		gists_url="https://api.github.com/users/user0/gists{/gist_id}",
		hireable=True,
		html_url="https://github.com/user0",
		id=109167,
		location=Here,
		login=user0,
		name="MyName",
		node_id="MDQ6VXNlcjEwOTE2Nw==",
		organizations_url="https://api.github.com/users/user0/orgs",
		owned_private_repos=73,
		plan="{'name': 'pro', 'space': 976562499, 'collaborators': 0, 'private_repos': 9999}",
		private_gists=2,
		public_gists=3,
		public_repos=38,
		received_events_url="https://api.github.com/users/user0/received_events",
		repos_url="https://api.github.com/users/user0/repos",
		site_admin=False,
		starred_url="https://api.github.com/users/user0/starred{/owner}{/repo}",
		subscriptions_url="https://api.github.com/users/user0/subscriptions",
		total_private_repos=7,
		twitter_username=user0,
		two_factor_authentication=True,
		type=User,
		updated_at="2024-04-15T20:40:59Z",
		url="https://api.github.com/users/user0",
		visit_started_at="2024-05-07T08:16:14Z"];
	"Users/user1"	[Type=ShadowUser,
		_id="Users/user1",
		_key=user1,
		_rev="_hzGd5G2---"];
	"Users/user1" -> "Users/user0"	[key=0,
		_from="Users/user1",
		_id="FOLLOW/user1-user0",
		_key="user1-user0",
		_rev="_hzGd5K----",
		_to="Users/user0"];
	"Users/user2"	[Type=ShadowUser,
		_id="Users/user2",
		_key=user2,
		_rev="_hzGd5NG---"];
	"Users/user2" -> "Users/user0"	[key=0,
		_from="Users/user2",
		_id="FOLLOW/user2-user0",
		_key="user2-user0",
		_rev="_hzGd5QO---",
		_to="Users/user0"];
	"Users/user3"	[Type=ShadowUser,
		_id="Users/user3",
		_key=user3,
		_rev="_hzGd5TW---"];
	"Users/user3" -> "Users/user0"	[key=0,
		_from="Users/user3",
		_id="FOLLOW/user3-user0",
		_key="user3-user0",
		_rev="_hzGd5XG---",
		_to="Users/user0"];
	"Users/user4"	[Type=ShadowUser,
		_id="Users/user4",
		_key=user4,
		_rev="_hzGd5aO---"];
}

I'm trying now to get a graphml file so I did

import networkx as nx
from networkx.drawing.nx_agraph import read_dot
nx_g = read_dot("gh.dot")
for node, attr_dict in nx_g.nodes(data=True):
    nx_g.nodes[node]["node-type"] = nx_g.nodes[node].pop("type", None)
nx.write_graphml_lxml(nx_g, "gh.graphml")

but it raises

TypeError: GraphML does not support type <class 'type'> as data values.

Any idea?

[scls19fr]

For better understanding here is full backtrace:

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
File ~\anaconda3\Lib\site-packages\networkx\readwrite\graphml.py:455, in GraphML.get_xml_type(self, key)
    454 try:
--> 455     return self.xml_type[key]
    456 except KeyError as err:

KeyError: <class 'dict'>

The above exception was the direct cause of the following exception:

TypeError                                 Traceback (most recent call last)
Cell In[25], line 1
----> 1 nx.write_graphml_lxml(nx_g, "gh.graphml")

File <class 'networkx.utils.decorators.argmap'> compilation 9:5, in argmap_write_graphml_lxml_5(G, path, encoding, prettyprint, infer_numeric_types, named_key_ids, edge_id_from_attribute)
      3 import gzip
      4 import inspect
----> 5 import itertools
      6 import re
      7 import warnings

File ~\anaconda3\Lib\site-packages\networkx\readwrite\graphml.py:176, in write_graphml_lxml(G, path, encoding, prettyprint, infer_numeric_types, named_key_ids, edge_id_from_attribute)
    165 except ImportError:
    166     return write_graphml_xml(
    167         G,
    168         path,
   (...)
    173         edge_id_from_attribute,
    174     )
--> 176 writer = GraphMLWriterLxml(
    177     path,
    178     graph=G,
    179     encoding=encoding,
    180     prettyprint=prettyprint,
    181     infer_numeric_types=infer_numeric_types,
    182     named_key_ids=named_key_ids,
    183     edge_id_from_attribute=edge_id_from_attribute,
    184 )
    185 writer.dump()

File ~\anaconda3\Lib\site-packages\networkx\readwrite\graphml.py:746, in GraphMLWriterLxml.__init__(self, path, graph, encoding, prettyprint, infer_numeric_types, named_key_ids, edge_id_from_attribute)
    743 self.attribute_types = defaultdict(set)
    745 if graph is not None:
--> 746     self.add_graph_element(graph)

File ~\anaconda3\Lib\site-packages\networkx\readwrite\graphml.py:780, in GraphMLWriterLxml.add_graph_element(self, G)
    778 for k, v in graphdata.items():
    779     print(k, v)  # scls
--> 780     element_type = self.get_xml_type(self.attr_type(k, "graph", v))
    781     self.get_key(str(k), element_type, "graph", None)
    782     print("ok")  # scls

File ~\anaconda3\Lib\site-packages\networkx\readwrite\graphml.py:457, in GraphML.get_xml_type(self, key)
    455     return self.xml_type[key]
    456 except KeyError as err:
--> 457     raise TypeError(
    458         f"GraphML does not support type {type(key)} as data values with {key}."  # modified by scls
    459     ) from err

TypeError: GraphML does not support type <class 'type'> as data values with <class 'dict'>.

I think it in fact caused by nx_g.graph["graph"] being a dict

nx_g.graph content is

{'name': 'VS-GH-Graph',
 'graph': {'name': 'VS-GH-Graph'},
 'node': {'Type': '',
  '_id': '',
  '_key': '',
  '_rev': '',
  'avatar_url': '',
  'bio': '',
  'blog': '',
  'collaborators': '',
  'company': '',
  'created_at': '',
  'disk_usage': '',
  'email': '',
  'events_url': '',
  'followers': '',
  'followers_url': '',
  'following': '',
  'following_url': '',
  'gists_url': '',
  'hireable': '',
  'html_url': '',
  'id': '',
  'location': '',
  'login': '',
  'name': 'VS-GH-Graph',
  'node_id': '',
  'organizations_url': '',
  'owned_private_repos': '',
  'plan': '',
  'private_gists': '',
  'public_gists': '',
  'public_repos': '',
  'received_events_url': '',
  'repos_url': '',
  'site_admin': '',
  'starred_url': '',
  'subscriptions_url': '',
  'total_private_repos': '',
  'twitter_username': '',
  'two_factor_authentication': '',
  'type': '',
  'updated_at': '',
  'url': '',
  'visit_started_at': ''},
 'edge': {'_from': '', '_id': '', '_key': '', '_rev': '', '_to': ''}}

[dschult]

To keep track of what is going on, keep in miind the picture of the data structure as dict-of-dicts-of.......
For node attributes:

• G.nodes is basically a dict keyed by node to a dict keyed by attribute names to the attribute value. So it's a dict-of-dicts.
• G.edges is basically a dict keyed by edge-tuple to a dict keyed by attribute names to the attribute value.

In Graph/DiGraph:

• G.adj is basically a dict keyed by node to a dict keyed by neighbors to a attr_dict keyed by attribute name to the attribute value.

In MultiGraph/MultiDiGraph:

• G.adj is basically a dict keyed by node to a dict keyed by neighbors to a dict keyed by edge_keys to a attr_dict keyed by attribute name to the attribute value.

Most of the bugs I see when looking at your code come from mistaking which level of the dict-of-dict-of-dict structure you are accessing. In addition, when you use set_node_attributes you need to supply a dict keyed by node. For set_edge_attributes you need to supply a dict keyed by edge-tuple. Keep in mind that an edge-tuple can be a 2-tuple or for multigraphs a 3-tuple.

Finally, graphml is fairly strict about what kinds of attribute values it can represent. You might try a different graph format if you want to use dicts or lists as the attribute values. (Though -- you should be able to use the name "type" for an attribute without trouble. The name of the attribute is not the problem -- it is the object type of the value that graphml is complaining about.