CODEWITHSUNDEEP
pythonpandasnetworkx
Aaaaad
Aaaaad

Reputation: 13

How to transform directed graph to undirected graph while recording information about attributes using NetworkX?

I am new to NetworkX and I have a problem that, I think, might be quite general: how can I take a directed network, convert it to an undirected network, and in this process record some information about the edges in the original directed network?

Specifically, I have a DiGraph in NetworkX that records links from id_from to id_to. Attributes for each edge are the month of the link and a weight.

I would like to convert this directed graph to an undirected graph where I record as attributes:

Here is an example of the pandas dataframe that I start with:

In [12]: df
Out[12]: 
    id_from     id_to     total       month
0         a         b     100.0  2014-01-01
1         b         a      10.0  2014-02-01
2         a         c      15.0  2014-01-01
3         c         d       7.0  2015-06-01
4         d         c     500.0  2016-03-01

I read this as a DiGraph:

In [13]: G = nx.from_pandas_dataframe(df, 'id_from', 'id_to', edge_attr = True, create_using = nx.DiGraph())
In [14]: print(G.edges(data = True))
Out[14]: [(a, b, {'id_from':  a, 'id_to': b, 'amount': 100.0, 'month': Timestamp('2014-01-01 00:00:00')}), (b, a, {'id_from':  b, 'id_to': a, 'amount': 10.0, 'month': Timestamp('2014-02-01 00:00:00')}), (a, c, {'id_from':  a, 'id_to': c, 'amount': 15.0, 'month': Timestamp('2014-01-01 00:00:00')}), (c, d, {'id_from':  c, 'id_to': d, 'amount': 7.0, 'month': Timestamp('2015-06-01 00:00:00')}), (d, c, {'id_from':  d, 'id_to': c, 'amount': 500.0, 'month': Timestamp('2016-03-01 00:00:00')})]

And then I would ultimately like to get back a graph, which I can then convert back into a pandas dataframe at some point, that looks like:

     id_one  id_two  total  first_month  last_month  nr_months  bidirect
0         a       b  110.0   2014-01-01  2014-02-01        2.0       Yes
1         a       c   15.0   2014-02-01  2014-02-01        1.0        No
2         c       d  507.0   2015-06-01  2016-03-01        2.0       Yes

Can anyone help me with this?

I can't seem to find any questions that are similar, but please correct me if I am wrong. Any help is much appreciated.

Upvotes: 1

Views: 2371

Answers (1)

zohar.kom
zohar.kom

Reputation: 1845

A possible approach is to use G.to_undirected() for transforming the directed graph to an undirected graph. Then iterating over the edges for updating the desired properties for each edge, and finally converting the graph to a dataframe:

import pandas as pd
import networkx as nx
import datetime

data = {
    'id_from': ['a', 'b', 'a', 'c', 'd'],
    'id_to': ['b', 'a', 'c', 'd', 'c'],
    'total': [100.0, 10.0, 15.0, 7.0, 500.0],
    'month': [datetime.datetime(2014, 1, 1), datetime.datetime(2014, 2, 1), datetime.datetime(2014, 1, 1), datetime.datetime(2015, 6, 1), datetime.datetime(2016, 3, 1)],
}
df = pd.DataFrame(data)
G = nx.from_pandas_edgelist(df, 'id_from', 'id_to', edge_attr=True, create_using=nx.DiGraph())

undirected = G.to_undirected()
for edge in undirected.edges(data=True):
    direction_1 = df.ix[(df['id_from'] == edge[0]) & (df['id_to'] == edge[1])]
    direction_2 = df.ix[(df['id_from'] == edge[1]) & (df['id_to'] == edge[0])]
    edges = pd.concat([direction_1, direction_2])
    edge[2]['bidirect'] = 'Yes' if (not direction_1.empty) & (not direction_2.empty) else 'No'
    edge[2]['total'] = edges['total'].sum()
    edge[2]['first_month'] = edges['month'].min()
    edge[2]['last_month'] = edges['month'].max()
    edge[2]['nr_months'] = edges['month'].nunique()
    del edge[2]['month']

print(nx.to_pandas_edgelist(undirected))

The result is:

  bidirect first_month last_month  nr_months source target  total
0       No  2014-01-01 2014-01-01          1      a      c   15.0
1      Yes  2014-01-01 2014-02-01          2      a      b  110.0
2      Yes  2015-06-01 2016-03-01          2      c      d  507.0

Each edge in a networkx graph, is essentially a tuple, where the first 2 elements are the edge's nodes and the last element (edge[2]) is a dictionary with the edge's properties. Hence, we can simply update this dictionary according to the desired logic.

Upvotes: 0

Related Questions