how to keep specific node in networkx

Question

I'm currently studying my college special topic. My problem is i can remove all the node i don't want but i want to keep some specific node. Here's how i do it.

1.read gml into networkx

2.using this code to remove the website that i don't want and then write it into a new gml file

import networkx as nx
G = nx.read_gml('test.gml')
for i in range(2000):
    for node in G.nodes:
        if "pu.edu.tw" not in node:
            G.remove_node(node)
            break
nx.write_gml(G,"finaltest.gml")

3.As you can see the part of this gml file, i successfully keep all 'pu.edu.tw' website

graph [
directed 1
multigraph 1
node [
  id 0
  label "https://www.pu.edu.tw/"
]
node [
  id 1
  label "https://alumni.pu.edu.tw/"
]
node [
  id 2
  label "https://freshman.pu.edu.tw/"
]
node [
  id 3
  label "https://tdc.pu.edu.tw/"
]
node [
  id 4
  label "https://alcat.pu.edu.tw/"
]
node [
  id 5
  label "https://www.secretary.pu.edu.tw/"
]
node [
  id 6
  label "https://pugive.pu.edu.tw/"
]

4.The problem is when i try to draw this gml file with networkx, i got some nodes without egdes

5.And i found out the reason is that I deleted the link related to 'pu.edu.tw' so there are some egdes missing

I want to know how to not only remove the website i don't want and keep specific node that related to 'pu.edu.tw' so that edges won't missing. or some way to reconnect node. Thank you.

---------------------------------------------------------------------------------

update a new question .... What if i want to add multiple condition, such as

def cleanup(g):
    g_aux = g.to_undirected()
        for node in g_aux.nodes:
            if ("tku.edu.tw"or"scu.edu.tw"or"cycu.edu.tw"or"fcu.edu.tw") not in node:
            for neighbor in g_aux.neighbors(node):
                if "tku.edu.tw"or"scu.edu.tw"or"cycu.edu.tw"or"fcu.edu.tw" in neighbor:
                    break
            else:
                g.remove_node(node)

is this the right way to do?

So, this orphans belong to 'pu.edu.tw', but were connected only to nodes that are not from this website, and you want to create edges that replace such indirect connections? Say, there were nodes 'A.pu.edu.tw' -- B.oth -- C.oth -- D.pu.edu.tw, then you want to add edge A -- C? — tyrrr, Dec 01 '20 at 13:39
Yes that is the way to do it if you want the nodes to have any of those ids — willcrack, Dec 22 '20 at 11:02

score 2 · Answer 1 · edited Dec 02 '20 at 16:26

If you only want to maintain one connection from each orphan node, the one with "closest" node from your subgraph, you may do the following: after creating subgraph, iterate over orphan nodes and for each of them perform BFS algorithm on the original graph, stopping when you find node that has label 'pw.edu.tw' and adding a new edge from this node to the orphan node in subgraph. With BFS you are guaranteed to find the closest node with desired property.

The following code should do the trick:

import networkx as nx
from networkx.algorithms.traversal.breadth_first_search import bfs_edges

G = nx.read_gml('test.gml')

desired_nodes = [node for node in G.nodes if 'pu.edu.tw' in node]
subgraph = nx.Graph(G.subgraph(desired_nodes))

orphan_nodes = [node for node in subgraph.nodes if 
subgraph.degree[node] == 0]

for orphan in orphan_nodes:
    for _, neigh in bfs_edges(G, orphan):
        if 'pu.edu.tw' in neigh:
            subgraph.add_edge(neigh, orphan)
            break
            
nx.write_gml(subgraph,"finaltest.gml")

I've also changed the method of removing nodes from graph - instead of the double loop you've implemented, I first find nodes with desired property with list comprahension, and then leverage the subgraph method from networkx.Graph - it's cleaner and will work for arbitrary number of removed nodes (in opposite to the loops, which you are probably aware of). This way a new graph object is created, instead of removing edges from an old one - which is necessary for algorithm presented above.

Also `G.subgraph(desired_nodes)` would return a message that you can't change a frozen graph, this is probably because they don't want you to take the risk of screwing a part of the original graph. Instead it should be `nx.Graph(G.subgraph(desired_nodes))`. Np ;) — willcrack, Dec 02 '20 at 16:46

willcrack · Accepted Answer · 2020-12-11T09:21:51.217

1

One thing you can do is to keep every node whose neighbor has "pu.edu.tw" in it's name.

Here's the full code:

import networkx as nx

def cleanup(g):
    g_aux = g.to_undirected()
    for node in g_aux.nodes:
        if "pu.edu.tw" not in node:
            for neighbor in g_aux.neighbors(node):
                if "pu.edu.tw" in neighbor:
                    # Found
                    break
            else:
                # Didn't find pu.edu.tw in any neighbors
                g.remove_node(node)

G = nx.read_gml('test.gml')
cleanup(G)
nx.write_gml(G,"finaltest.gml")

The result obtained is every node with "pu.edu.tw" and it's neighbors.
Please note that I used an undirected version of the graph, g_aux = g.to_undirected(), keeping every neighbor of a "pu.edu.tw" independently of the direction of the connecting edge.

Here is some code to check if any pu.edu.tw doesn’t have any neighbors:

def check_isolated(g):
    for node in g.nodes:
        if "pu.edu.tw" in node:
            if g.degree[node] == 0:
                print(node)

If this outputs anything before running cleanup then those nodes will always be isolated.

print(“before”)
check_isolated(g)
print(“cleaning...”)
cleanup(g)
print(“after”)
check_isolated(g)

edited Dec 11 '20 at 09:21

answered Dec 02 '20 at 11:39

willcrack

1,794
11
20

Wouldn't it miss cases when nodes with `'pw.edu.tw'` are connected via more then 2 nodes without this label? – tyrrr Dec 02 '20 at 12:41
Yess it would. It just keeps in the graph every node whose neighbor has "pu.edu.tw" in it's name. – willcrack Dec 02 '20 at 15:40
@willcrack i'm sorry but there is still a few node that isolated to the others. – KaiHung Dec 10 '20 at 08:45
Have you been able to identify why? – willcrack Dec 10 '20 at 11:12
Have you checked if those isolated nodes weren’t isolated in the original graph? – willcrack Dec 10 '20 at 19:30
I’ve added some code for you to check those nodes – willcrack Dec 10 '20 at 19:41
Thanks for your reply. The code that you gave me doesn't print any node. Or can you tell me where should i put the code exactly? – KaiHung Dec 11 '20 at 05:33
I’ve added some code explaining what I mean – willcrack Dec 11 '20 at 09:22
After reworking my GML file, there is no isolated node at all. – KaiHung Dec 13 '20 at 09:10
1.what is `g_aux` mean? 2. so i have to change `undirected` into `directed` if my graph is Multigrpah? – KaiHung Dec 15 '20 at 10:26
1. Its supposed to mean graph auxiliary. \\ 2. I don't think you have to change anything if you are using a multigraph – willcrack Dec 15 '20 at 11:30
I'm sorry but can you comment what each code mean? Thank you :) – KaiHung Dec 20 '20 at 03:26
And when i change undirected into directed, the result changed.whys that – KaiHung Dec 20 '20 at 06:23
Could you post a new question with the details? – willcrack Dec 20 '20 at 11:03
i added a new question below – KaiHung Dec 21 '20 at 15:00

KaiHung · Answer 3 · 2020-12-21T14:58:03.597

0

Here is a question i want to ask....

g_aux = g.to_undirected()

why do i have to use g_aux to run this program? i don't understand what graph auxiliary really do in NetworkX.

edited Dec 21 '20 at 14:58

answered Dec 20 '20 at 16:23

KaiHung

23
5

`g_aux` is just a variable, it is an undirected version of the graph `g` – willcrack Dec 21 '20 at 15:05
Also, it would be preferable for you to add an edit to your question than to ask a question on an answer post. – willcrack Dec 21 '20 at 15:07
ok, i updated an new question in my original post please check that out.... – KaiHung Dec 22 '20 at 09:53
I'm sorry, icant find it, I can you post the link here? – willcrack Dec 22 '20 at 10:47
just check my original question, i added an update below – KaiHung Dec 22 '20 at 10:52

how to keep specific node in networkx

3 Answers3