KeyError when assign colours to nodes

Question

I am getting a KeyError when I try to create a network.

My dataset is

Node    Neighbors       Colour  Weight
 Luke   Alte            orange    3
 Luke   John            orange    3
Michael Laura           red       43
Ludo    Stella          orange   21
Alte    Ludo            blue     24
Alte    Luke            blue     24

The table above shows the links by nodes:

node Luke is linked with Alte and John. It has edge weight 3 and colour orange
node Michael is linked with Laura. It has weight 43 and colour red
node Ludo is linked with Stella. It has weight 21 and colour orange
node Alte is linked with Luke and Ludo. It has colour blue and weight 24

Doing as follows:

NROWS = None
def get_graph_from_pandas(df):
    
    G = nx.DiGraph() # assuming the graph is directed since e.g node 1 has 
                     # 3 as neighbour but 3 doesnt have 1 as neighbour
    
    
    for row in df.itertuples(): # row is the row of the dataframe
        n = row.Node
        w = row.Weight
        c = row.Colour
        neighbors = row.Neighbors
        
        G.add_node(n, weight = w, colour = c)
        
        for neigh in neighbors:
            #add edge weights here, attribute of G.add_edge
            G.add_edge(n,neigh)  
            
    return G
        
        
        
G = get_graph_from_pandas(df)

print("Done.")
print("Total number of nodes: ", G.number_of_nodes())
print("Total number of edges: ", G.number_of_edges())

pos = nx.draw(G, with_labels=True, 
              node_color=[node[1]['colour'] for node in G.nodes(data=True)], 
              node_size=200)

gives me a KeyError: 'colour'.

When I print

for node in G.nodes(data=True):     
     try:         
        node[1]['colour']     
     except KeyError:         
        print(node)

I get

('A', {}) 
('l', {}) 
('t', {}) 
('e', {})

Can you please explain what is causing the error? Thanks

Update: I think the error is from here

 for neigh in neighbors:
                #add edge weights here, attribute of G.add_edge
                G.add_edge(n,neigh)

It's not just a case sensitivity issue is it? i.e. you have `Colour` in your example but your code uses `colour`. — Random Davis, Dec 08 '20 at 19:25
I already tried with Colour but it gives the same error :( I updated the question to show the how nodes look like (as you can see it is completely wrong, as it should be Luke and not L-u-k-e-) — V_sqrt, Dec 08 '20 at 19:26
When you [catch the error](https://docs.python.org/3/tutorial/errors.html#handling-exceptions) and inspect/print relevant data in the except suite do you see anything unusual? - is everything as expected? If you are using an IDE **now** is a good time to learn its debugging features Or the built-in [Python debugger](https://docs.python.org/3/library/pdb.html). ... [What is a debugger and how can it help me diagnose problems?](https://stackoverflow.com/questions/25385173/what-is-a-debugger-and-how-can-it-help-me-diagnose-problems) — wwii, Dec 08 '20 at 19:35
I can see that the other letters that print are not in order. For example, `('L', {}) ('u', {}) ('k', {}) ('e', {}) ('M',{}), ('h',{})`. I think the problem is that is catching only the first letter of the words (splitting it?) — V_sqrt, Dec 08 '20 at 19:37
Your for loop over Neighbors loops over the characters in the "Neighbor" item and adds it as an end-node in the edge, instead of the actual Neighbor. And since there isn't a node with that name, networkx adds that as a node in your graph. You should remove that for loop and replace it with `G.add_edge(n,neighbors)` — cookesd, Dec 08 '20 at 20:20

wwii · Answer 1 · 2020-12-09T17:09:01.583

Each item in df.Neighbors is a string. When you iterate over it with for neigh in neighbors: You add each character of the neighbor to the node. For example the first node looks like

>>> G.nodes
>>> NodeView(('Luke', 'A', 'l', 't', 'e'))

As long as each row only has a single Neighbor, replace the for loop with

    # for neigh in neighbors:
    #     #add edge weights here, attribute of G.add_edge
    #     G.add_edge(n,neigh)  
    G.add_edge(n,neighbors)

Although this doesn't alleviate the KeyError.

While 'John', 'Laura', and 'Stella' are neighbors they are also nodes in the graph but they were created with .add_edge and never had a color assigned to them.

>>> for thing in G.nodes.items():
...     print(thing)
('Luke', {'weight': 3, 'colour': 'orange'})
('Alte', {'weight': 24, 'colour': 'blue'})
('John', {})
('Michael', {'weight': 43, 'colour': 'red'})
('Laura', {})
('Ludo', {'weight': 21, 'colour': 'orange'})
('Stella', {})

You can add those nodes first with default attributes before iterating:

...
    G.add_nodes_from(df.Neighbors,colour='white',weight=0)
    for row in df.itertuples(): # row is the row of the dataframe
        ...

If your node attributes can begin with capitals the graph construction could be written:

def get_graph_from_pandas(df):
    
    G = nx.DiGraph() # assuming the graph is directed since e.g node 1 has 
                     # 3 as neighbour but 3 doesnt have 1 as neighbour
    
    
    G.add_nodes_from(df.Neighbors,Colour='white',Weight=0)
    G.add_edges_from(df[['Node','Neighbors']].itertuples(index=False))
    dg = df.set_index('Node')
    G.add_nodes_from(dg[['Colour','Weight']].T.to_dict().items())
        
    return G

>>> for thing in G.nodes(data=True):
...     print(thing)
('Alte', {'Colour': 'blue', 'Weight': 24})
('John', {'Colour': 'white', 'Weight': 0})
('Laura', {'Colour': 'white', 'Weight': 0})
('Stella', {'Colour': 'white', 'Weight': 0})
('Ludo', {'Colour': 'orange', 'Weight': 21})
('Luke', {'Colour': 'orange', 'Weight': 3})
('Michael', {'Colour': 'red', 'Weight': 43})
>>> for thing in G.edges(data=True):
...     print(thing)
('Alte', 'Ludo', {})
('Alte', 'Luke', {})
('Ludo', 'Stella', {})
('Luke', 'Alte', {})
('Luke', 'John', {})
('Michael', 'Laura', {})

You can get the node colors directly from G.nodes.items

pos = nx.draw(G, with_labels=True, 
              node_color=[d['Colour'] for n,d in G.nodes.items()], 
              node_size=200)

or nx.get_node_attributes

pos = nx.draw(G, with_labels=True, 
              node_color=nx.get_node_attributes(G,'Colour').values(),
              node_size=200)

thanks wwiii. Unfortunately I am still getting the same error. The G.nodes prints correctly the nodes. However, I am getting the same error for colour/Colour. May I ask you if you get similar error? (if you can check it with the data I provided). — V_sqrt, Dec 08 '20 at 20:19

willcrack · Accepted Answer · 2020-12-08T21:01:59.020

wwii answer solves one problem.

However there are a number of problems that need to be fixed:

Only nodes in column Node will have color, users that are only introduced in Neighbors column will be created in G.add_edge(n,neighbor), and won't have a color assigned. You need to decide which color to set for these nodes.
The weight you want to attribute to the edges is being attributed to the nodes.

df = pd.DataFrame(  data = {"Node": ["Luke", "Luke", "Michael", "Ludo", "Alte", "Alte"],
                            "Neighbors": ["Ludo", "John", "Laura", "Stella", "Ludo", "Luke"],
                            "Colour": ["orange", "orange", "red", "orange", "blue", "blue"], 
                            "Weight": [3, 3 ,43, 21, 24, 24] 
                        }
              )
   

NROWS = None
def get_graph_from_pandas(df, v = False):
    
    G = nx.DiGraph() # assuming the graph is directed since e.g node 1 has 
                     # 3 as neighbour but 3 doesnt have 1 as neighbour
    
    for row in df.itertuples():
        print(row)
        n = row.Node
        w = row.Weight
        c = row.Colour
        neighbor = row.Neighbors
        
        G.add_node(n, weight = w, colour = c) # only nodes in column Node will have color
                                              # users that are only introduced in Neighbors column dwont have column
        if neighbor not in G.nodes:
            G.add_node(neighbor, weight = w, colour = "yellow") # this will set the default color to yellow
        G.add_edge(n,neighbor, weight = w) # weight of edge
            
    return G
        
G = get_graph_from_pandas(df, v = False)

print("Done.")
print("Total number of nodes: ", graph.number_of_nodes())
print("Total number of edges: ", graph.number_of_edges())

fig = plt.figure(figsize=(2,2))

pos = nx.draw(G, with_labels=True, 
              node_color=[node[1]['colour'] for node in G.nodes(data=True)], 
              node_size=200)

for node in G.nodes(data=True):
    try:
        node[1]['colour']
    except KeyError:
        print(node)

thanks willcrack. So the colour in the Colour column is the colour of nodes in Node column. All of them should have a colour. If they do not have because, for example, they are not in the Nodes column, then I would assign a default colour (e.g. yellow), just to avoid this potential issue — V_sqrt, Dec 08 '20 at 20:23
This was happening because your Neighbors column used to be a column of lists — willcrack, Dec 08 '20 at 20:29
Also, I added an edit. You should only change a node to yellow if the node is not already in the graph — willcrack, Dec 08 '20 at 20:30

KeyError when assign colours to nodes

2 Answers2

Linked