Dependency trees to DGL graphs

Questions
#1

Hello,

Do you have any example on how to convert dependency trees (from spacy or stanford corenlp) into DGL graphs that can serve as input to your TreeLSTM model?

Thanks in advance!

#2

We don’t have the dataset in DGL currently. Welcome contribution! I guess if you could first convert it to networkx graph, then it can be converted to DGLGraph easily.

#3 
nlp = spacy.load("en_core_web_sm")    
doc = nlp(u'The last reply to this topic was 10 months ago.')

Load spacy’s dependency tree into a networkx graph

edges = []
for token in doc:
    for child in token.children:
        edges.append(('{0}'.format(token.lower_),
                      '{0}'.format(child.lower_)))

visualize the dependency tree in google colab

spacy.displacy.render(doc, style='dep', jupyter=True)

convert networkx graph to dgl graph

graph = nx.Graph(edges)
g = dgl.DGLGraph()
g.from_networkx(graph)

function to visualize the graph created

def plot_tree(g):
    pos = nx.nx_agraph.graphviz_layout(g, prog='dot')
    nx.draw(g, pos, with_labels=False, node_size=10,
            node_color=[[.5, .5, .5]], arrowsize=4)
    plt.show()

Visualize dgl graph as a networkx graph

plot_tree(g.to_networkx())

2 Likes
#4

I have written a python script to create a DGL graph object using a spacy dependency parser, unlike the above answers it doesn’t require the networkx package.

2 Likes