Best practices for Querying graphs by edge and node attributes in NetworkX

Question

Using NetworkX, and new to the library, for a social network analysis query. By Query, I mean select/create subgraphs by attributes of both edges nodes where the edges create a path, and nodes contain attributes. The graph is using a MultiDiGraph of the form

G2 = nx.MultiDiGraph()
G2.add_node( "UserA", { "type" :"Cat" } )
G2.add_node( "UserB", { "type" :"Dog" } )
G2.add_node( "UserC", { "type" :"Mouse" } )
G2.add_node( "Likes", { "type" :"Feeling" } )
G2.add_node( "Hates", { "type" :"Feeling" } )

G2.add_edge( "UserA", 'Hates' ,  statementid="1" )
G2.add_edge( "Hates", 'UserB' ,  statementid="1"  )
G2.add_edge( "UserC", 'Hates' ,  statementid="2" )
G2.add_edge( "Hates", 'UserA' ,  statementid="2"  )
G2.add_edge( "UserB", 'Hates' ,  statementid="3"  )
G2.add_edge( "Hates", 'UserA' ,  statementid="3"  )
G2.add_edge( "UserC", 'Likes' ,  statementid="3"  )
G2.add_edge( "Likes", 'UserB' ,  statementid="3"  )

Queried with

for node,data in G2.nodes_iter(data=True):
    if ( data['type'] == "Cat" ):
       # get all edges out from these nodes
            #then recursively follow using a filter for a specific statement_id

#or get all edges with a specific statement id
   # look for  with a node attribute of "cat" 

Is there a better way to query? Or is it best practice to create custom iterations to create subgraphs?

Alternatively (and a separate question), the Graph could be simplified, but I'm not using the below graph because the "hates" type objects will have predcessors. Would this make querying simpler? Seems easier to iterate over nodes

G3 = nx.MultiDiGraph()
G3.add_node( "UserA", { "type" :"Cat" } )
G3.add_node( "UserB", { "type" :"Dog" } )

G3.add_edge( "UserA", 'UserB' ,  statementid="1" , label="hates")
G3.add_edge( "UserA", 'UserB' ,  statementid="2" , label="hates")

Other notes:

• Perhaps add_path adds an identifier to the path created?
• iGraph has a nice query feature g.vs.select()

Answer 1

It's pretty straightforward to write a one-liner to make a list or generator of nodes with a specific property (generators shown here)

import networkx as nx

G = nx.Graph()
G.add_node(1, label='one')
G.add_node(2, label='fish')
G.add_node(3, label='two')
G.add_node(4, label='fish')

# method 1
fish = (n for n in G if G.node[n]['label']=='fish')
# method 2
fish2 = (n for n,d in G.nodes(data=True) if d['label']=='fish')

print(list(fish))
print(list(fish2))

G.add_edge(1,2,color='red')
G.add_edge(2,3,color='blue')

red = ((u,v) for u,v,d in G.edges(data=True) if d['color']=='red')

print(list(red))

If your graph is large and fixed and you want to do fast lookups you could make a "reverse dictionary" of the attributes like this,

labels = {}
for n, d in G.nodes(data=True):
    l = d['label']
    labels[l] = labels.get(l, [])
    labels[l].append(n)
print labels

Answer 2

Building on @Aric's answer, you can find red fish like this:

red_fish = set(n for u,v,d in G.edges_iter(data=True)
               if d['color']=='red'
               for n in (u, v)
               if G.node[n]['label']=='fish')

print(red_fish)
# set([2])

Answer 3

In order to select edges based on attributes of edges AND nodes, you may want to do something like this, using your graph, G2:

def select(G2, query):
    '''Call the query for each edge, return list of matches'''
    result = []
    for u,v,d in G2.edges(data=True):
        if query(u,v,d):
            result.append([(u,v)])
    return result

# Example query functions
# Each assumes that it receives two nodes (u,v) and 
# the data (d) for an edge 

def dog_feeling(u, v, d):
    return (d['statementid'] == "3" 
            and G2.node[u]['type'] == "Dog"
            or G2.node[u]['type'] == "Dog")

def any_feeling(u,v,d):
    return (d['statementid'] == "3" 
            and G2.node[u]['type'] == "Feeling"
            or G2.node[u]['type'] == "Feeling")

def cat_feeling(u,v,d):
    return (G2.node[u]['type'] == "Cat"
            or G2.node[v]['type'] == "Cat")

# Using the queries
print select(G2, query = dog_feeling)
print select(G2, query = any_feeling)
print select(G2, query = cat_feeling)

This abstracts away the iteration process into the select() function and you can write your queries as individual, testable functions.