def ego_graph(self, radius=1, types=None, min_degree=None):
'''Generate an undirected ego graph around the current entity.
:param radius: radius or degree of the ego graph; defaults to 1
:param types: node types to be included in the graph (e.g., restrict
to people and organizations only)
:param min_degree: optionally filter nodes in the generated ego graph
by minimum degree
'''
network = network_data()
undirected_net = network.to_undirected()
# filter network *before* generating ego graph
# so we don't get disconnected nodes
if types is not None:
for n in undirected_net.nodes():
if 'type' not in undirected_net.node[n] or \
undirected_net.node[n]['type'] not in types:
undirected_net.remove_node(n)
# converted multidigraph to undirected
# to make it possible to find all neighbors,
# not just outbound connections
# (should be a way to get this from a digraph...)
eg = nx.ego_graph(undirected_net, self.nx_node_id,
radius=radius)
if min_degree is not None:
return filter_graph(eg, min_degree=min_degree)
return eg
def group_people_js(request, mode="egograph", output="full"):
"""Return Belfast Group network graph data as JSON, for use with
:meth:`group_people`.
Optionally filter
the data by minimum degree, if min_degree is specified as a url parameter.
When generating the node and link data, nodes are annotated with
degree, in degree, out degree, betweenness centrality, and eigenvector
centrality if available. (In/out degree is only available for directed
graphs.)
:param mode: egograph: network information for a one- or two-degree
egograph centered around the Belfast Group; groupsheet-model:
alternate network graph based on the Group sheets themselves
:param output: full: node and link data; adjacency: adjacency matrix,
used for generating chord diagram
"""
if mode == "egograph":
degree = request.GET.get("degree", 1)
extra_opts = {}
try:
degree = int(degree)
# currently only support 1 or 2 degree
degree = max(1, min(degree, 2))
# NOTE: degree 2 graph is large enough that it *must* be filtered
# to be sensible and usable on the webpage;
# by trial & error, I found a minimum degree of 5 to be reasonable
if degree == 2:
extra_opts["min_degree"] = 5
except ValueError:
# if a value is passed that can't be converted to int, fallback to 1
degree = 1
belfast_group = RdfOrganization(network_data().copy(), BELFAST_GROUP_URI)
graph = belfast_group.ego_graph(radius=degree, types=["Person", "Organization"], **extra_opts)
# annotate nodes in graph with degree
# FIXME: not a directional graph; in/out degree not available
graph = annotate_graph(
graph, fields=["degree", "in_degree", "out_degree", "betweenness_centrality", "eigenvector_centrality"]
)
elif mode == "groupsheet-model":
graph = gexf.read_gexf(settings.GEXF_DATA["bg1"])
graph = annotate_graph(
graph, fields=["degree", "betweenness_centrality", "eigenvector_centrality"] #'in_degree', 'out_degree',
)
if output == "full":
data = json_graph.node_link_data(graph)
if output == "adjacency":
# adjacency matrix for generating chord diagram
matrix = nx.convert_matrix.to_numpy_matrix(graph)
# NOTE: this also works, but as of networx 1.9 requires scipy
# matrix = nx.linalg.graphmatrix.adjacency_matrix(graph)
data = matrix.tolist()
return HttpResponse(json.dumps(data), content_type="application/json")
def connections(self, rdftype=None, resource=rdflib.resource.Resource):
'''Generate a dictionary of connected entities (direct neighbors
in the network graph) with a list of relationship terms (edge labels).
Optionally, takes an RDF type to filter the entities (e.g., restrict
only to People or Organizations), and a subclass of
:class:`rdflib.resource.Resource` to initialize the entity as.'''
network = network_data()
graph = rdf_data()
if self.nx_node_id not in network.nodes():
return {}
# this also works...
# neighbors = network.neighbors(self.nx_node_id)
ego_graph = self.ego_graph()
neighbors = ego_graph.nodes()
connections = {}
for node in neighbors:
weight = 0
# don't include the current person in their own connections
if node == self.nx_node_id:
continue
uriref = rdflib.URIRef(node)
# if an rdf type was specified, filter out items that do not
# match that type.
if rdftype is not None and \
(uriref, rdflib.RDF.type, rdftype) not in graph:
continue
res = resource(graph, uriref)
rels = set()
# find any edges between this node and me
# include data to simplify accessing edge label
# use edges & labels from original multidigraph
all_edges = network.out_edges(node, data=True) + \
network.in_edges(node, data=True)
for edge in all_edges:
src, target, data = edge
if self.nx_node_id in edge and 'label' in data:
weight += data.get('weight', 1) # assume default of 1 if not set
rels.add(data['label'])
if 'knows' in rels and 'correspondedWith' in rels:
rels.remove('knows')
# connections[res] = {'rels': rels, 'weight': weight}
connections[res] = (rels, weight)
# sort by weight so strongest connections will be listed first
conn = sorted(connections.items(), key=lambda x: x[1][1], reverse=True)
return conn
def gexf_content(request, mode):
"""Make network data available as GEXF files for download and use in
tools like Gephi."""
if mode == "all":
graph = network_data()
elif mode == "group-people":
# filtered graph of people/places/organizations used for
# first BG network graph displayed on the site
# - same data used in :meth:`full_js`
graph = _network_graph()
elif mode == "groupsheets":
graph = gexf.read_gexf(settings.GEXF_DATA["bg1"])
buf = StringIO()
gexf.write_gexf(graph, buf)
response = HttpResponse(buf.getvalue(), content_type="application/gexf+xml")
response["Content-Disposition"] = "attachment; filename=belfastgroup-%s.gexf" % mode
return response
def _network_graph(min_degree=1, **kwargs):
graph = network_data().copy() # don't modify the original network
rdfgraph = rdf_data()
# filter graph by type of node
types = ["Person", "Organization", "Place", "BelfastGroupSheet"]
for n in graph.nodes():
if "type" not in graph.node[n] or graph.node[n]["type"] not in types:
graph.remove_node(n)
continue
# use groupsheets to infer a connection between the author
# of the groupsheet and the group itself
# FIXME: this needs to be in data prep/clean, NOT here
# TODO: should be handled in prep now; confirm and then remove this logic
if graph.node[n]["type"] == "BelfastGroupSheet":
sheet = RdfGroupSheet(rdfgraph, rdflib.URIRef(n))
# FIXME: error handling when author is not in the graph?
# should probably at least log this...
if sheet.author and unicode(sheet.author.identifier) in graph:
graph.add_edge(unicode(sheet.author.identifier), BELFAST_GROUP_URI, weight=4)
# remove the groupsheet itself from the network, to avoid
# cluttering up the graph with too much information
# graph.add_edge(n, BELFAST_GROUP_URI, weight=5)
graph.remove_node(n)
# AFTER filtering by type, filter out by requested minimum degree
removed = 0
for n in graph.nodes():
if graph.degree(n) < min_degree:
removed += 1
graph.remove_node(n)
logger.info("removed %d nodes with degree less than %d" % (removed, min_degree))
return graph
