CODEWITHSUNDEEP
pythonjsontreenetworkxgraphviz
JoseOrtiz3
JoseOrtiz3

Reputation: 1883

Convert graphviz.dot.Digraph to networkx.Graph

Question

How do I convert a graphviz.dot.Digraph into a networkx.Graph (or any subclass of it)?

Motivation

LightGBM, an implementation of a tree-based algorithm, has a function returning a graphviz.dot.Digraph object. This type of object can represent any directed graph, but my graph is specifically a tree, and thus can be represented in JSON via a simpler nested structure:

var tree = {
    "name": "parent",
    "children": [
         {
         "name": "child_1"
         }
         {
         "name": "child_2"
         }
    [
}

Another, longer example of the above JSON structure is here. I use this JSON format to create a tree visualization using d3 in javascript.

In total, I need to convert this graphviz.dot.Digraph object to the above nested-JSON format.

If I can convert this graphviz.dot.Digraph object into a networkx.Graph object, I can use this method to convert it to the required JSON format. This intermediate conversion has been problematic for me. It seems I need another conversion to something networkx can use.

Upvotes: 2

Views: 3741

Answers (1)

JoseOrtiz3
JoseOrtiz3

Reputation: 1883

One method of converting graphviz.dot.Digraph to a networkx.Graph is to convert it to a pydotplus.graphviz.Dot object, and convert that into a networkx.classes.multidigraph.MultiDiGraph object which inherets from graph.

#--------------------
# Do whatever it is you do to create the graphviz.Digraph object
import lightgbm as lgbm

# .......
digraph = lgbm.create_tree_digraph(model)
print(type(digraph)) # prints <class 'graphviz.dot.Digraph'>


#--------------------
# Perform the conversion to networkx
import pydotplus
import networkx

dotplus = pydotplus.graph_from_dot_data(digraph.source)
print(type(dotplus)) # prints <class 'pydotplus.graphviz.Dot'>
# if graph doesn't have multiedges, use dotplus.set_strict(true)
nx_graph = networkx.nx_pydot.from_pydot(dotplus)
print(type(nx_graph)) # prints <class 'networkx.classes.multidigraph.MultiDiGraph'>

The rest is just for people considering conversion to JSON

#--------------------
# Optionally, convert to json
import json

# must specify whatever node is the root of the tree, here its 'split0'
#     or you can try list(graph)[0] if you're not sure.
data = networkx.readwrite.json_graph.tree_data(nx_graph,root='split0')
# data is just a dictionary. Dump results in proper json format.
json.dump(data,open(file_path,'w'))

Later I found out LightGBM actually has a dump_model method that shortcuts all of this...but I'll leave this here.

Upvotes: 1

Related Questions