How to label edges and avoid the edge overlapping in MultiDiGraph/DiGraph? (Networkx)

Question

Here is my code now

G = nx.from_pandas_edgelist(data, source='grad',
                                target='to', edge_attr='count',
                                create_using=nx.DiGraph())
weight = nx.get_edge_attributes(G, 'count')
pos = nx.shell_layout(G, scale=1)
nx.draw_networkx_nodes(G, pos, node_size=300, node_color='lightblue')
nx.draw_networkx_labels(G, pos=pos, font_color='red')
nx.draw_networkx_edges(G, pos=pos, edgelist=G.edges(), edge_color='black',
                       connectionstyle='arc3, rad = 0.1')
nx.draw_networkx_edge_labels(G, pos=pos, edge_labels=weight)
plt.show()

and the Result plot

serverl problems in this plot:

edge labels is not fully displayed,some edge doesn't have a label
edge labels are left on their own edge since I add the curve to the edge to avoid the edge overlapping
I think the edges which are not bidirectional don't need the curve, how to show them more neatly?

The data.head(50).to_dict():

{'grad': {0: 'CUHK', 1: 'CUHK', 2: 'CUHK', 3: 'CUHK', 4: 'CUHK', 5: 'CUHK', 6: 'CUHK', 7: 'CUHK', 8: 'CityU', 9: 'CityU', 10: 'CityU', 11: 'CityU', 12: 'CityU', 13: 'CityU', 14: 'CityU', 15: 'HKBU', 16: 'HKU', 17: 'HKU', 18: 'HKU', 19: 'HKU', 20: 'HKU', 21: 'HKU', 22: 'HKU', 23: 'HKUST', 24: 'HKUST', 25: 'HKUST', 26: 'HKUST', 27: 'HKUST', 28: 'HKUST', 29: 'HKUST', 30: 'HKUST', 31: 'Low Frequency', 32: 'Low Frequency', 33: 'Low Frequency', 34: 'Low Frequency', 35: 'Low Frequency', 36: 'Low Frequency', 37: 'Low Frequency', 38: 'Low Frequency', 39: 'PolyU', 40: 'PolyU', 41: 'PolyU', 42: 'PolyU', 43: 'PolyU', 44: 'PolyU'}, 'to': {0: 'CUHK', 1: 'CityU', 2: 'EduHK', 3: 'HKBU', 4: 'HKU', 5: 'HKUST', 6: 'LingU', 7: 'PolyU', 8: 'CityU', 9: 'EduHK', 10: 'HKBU', 11: 'HKU', 12: 'HKUST', 13: 'LingU', 14: 'PolyU', 15: 'HKBU', 16: 'CUHK', 17: 'CityU', 18: 'EduHK', 19: 'HKBU', 20: 'HKU', 21: 'HKUST', 22: 'PolyU', 23: 'CUHK', 24: 'CityU', 25: 'EduHK', 26: 'HKBU', 27: 'HKU', 28: 'HKUST', 29: 'LingU', 30: 'PolyU', 31: 'CUHK', 32: 'CityU', 33: 'EduHK', 34: 'HKBU', 35: 'HKU', 36: 'HKUST', 37: 'LingU', 38: 'PolyU', 39: 'CityU', 40: 'EduHK', 41: 'HKBU', 42: 'HKU', 43: 'LingU', 44: 'PolyU'}, 'count': {0: 13, 1: 6, 2: 3, 3: 6, 4: 5, 5: 3, 6: 2, 7: 6, 8: 4, 9: 1, 10: 5, 11: 2, 12: 1, 13: 2, 14: 7, 15: 2, 16: 2, 17: 4, 18: 3, 19: 1, 20: 17, 21: 3, 22: 9, 23: 4, 24: 2, 25: 2, 26: 4, 27: 2, 28: 4, 29: 4, 30: 6, 31: 76, 32: 73, 33: 1, 34: 16, 35: 57, 36: 46, 37: 3, 38: 69, 39: 1, 40: 2, 41: 3, 42: 1, 43: 1, 44: 23}}

Regarding points 1 and 2, you should be able to use the `offset` function for moving labels that I define [here](https://stackoverflow.com/a/74480623/2476977). — Ben Grossmann, Nov 29 '22 at 18:53
It would be helpful if you could share the data so that we can reproduce the graph on our end. Perhaps you can edit your post and paste in the result of `data.head().to_dict()` — Ben Grossmann, Nov 29 '22 at 19:32
Regarding 3, there's no way to separately set the connection styles for the edges within the graph using networkx; you'd need some other module or to draw the edges using pyplot directly. — Ben Grossmann, Nov 29 '22 at 19:40
I have pasted my data below the question, thank you if you can help with this. I'm still trying, the row"Low Frequency" can be removed if the plot is too messy. — kenny, Nov 29 '22 at 19:49
Also, there have some self-loop edges doesn't be displayed. It would be great if that could also be displayed. — kenny, Nov 29 '22 at 20:05
Interesting; when I run your code, I get self-loops. What do you have for `nx.__version__`? mine is 2.6.3. — Ben Grossmann, Nov 29 '22 at 20:08
Nevermind, my version is 2.4, it can be shown now after I upgrade it! — kenny, Nov 30 '22 at 02:59

Ben Grossmann · Accepted Answer · 2022-11-30T13:07:58.663

Here's a solution to issues 1 and 2. In my version of networkx, self-loops are displayed.

import pandas as pd
import networkx as nx
import numpy as np
import matplotlib.pyplot as plt

d = {'grad': {0: 'CUHK', 1: 'CUHK', 2: 'CUHK', 3: 'CUHK', 4: 'CUHK', 5: 'CUHK', 6: 'CUHK', 7: 'CUHK', 8: 'CityU', 9: 'CityU', 10: 'CityU', 11: 'CityU', 12: 'CityU', 13: 'CityU', 14: 'CityU', 15: 'HKBU', 16: 'HKU', 17: 'HKU', 18: 'HKU', 19: 'HKU', 20: 'HKU', 21: 'HKU', 22: 'HKU', 23: 'HKUST', 24: 'HKUST', 25: 'HKUST', 26: 'HKUST', 27: 'HKUST', 28: 'HKUST', 29: 'HKUST', 30: 'HKUST', 31: 'Low Frequency', 32: 'Low Frequency', 33: 'Low Frequency', 34: 'Low Frequency', 35: 'Low Frequency', 36: 'Low Frequency', 37: 'Low Frequency', 38: 'Low Frequency', 39: 'PolyU', 40: 'PolyU', 41: 'PolyU', 42: 'PolyU', 43: 'PolyU', 44: 'PolyU'}, 'to': {0: 'CUHK', 1: 'CityU', 2: 'EduHK', 3: 'HKBU', 4: 'HKU', 5: 'HKUST', 6: 'LingU', 7: 'PolyU', 8: 'CityU', 9: 'EduHK', 10: 'HKBU', 11: 'HKU', 12: 'HKUST', 13: 'LingU', 14: 'PolyU', 15: 'HKBU', 16: 'CUHK', 17: 'CityU', 18: 'EduHK', 19: 'HKBU', 20: 'HKU', 21: 'HKUST', 22: 'PolyU', 23: 'CUHK', 24: 'CityU', 25: 'EduHK', 26: 'HKBU', 27: 'HKU', 28: 'HKUST', 29: 'LingU', 30: 'PolyU', 31: 'CUHK', 32: 'CityU', 33: 'EduHK', 34: 'HKBU', 35: 'HKU', 36: 'HKUST', 37: 'LingU', 38: 'PolyU', 39: 'CityU', 40: 'EduHK', 41: 'HKBU', 42: 'HKU', 43: 'LingU', 44: 'PolyU'}, 'count': {0: 13, 1: 6, 2: 3, 3: 6, 4: 5, 5: 3, 6: 2, 7: 6, 8: 4, 9: 1, 10: 5, 11: 2, 12: 1, 13: 2, 14: 7, 15: 2, 16: 2, 17: 4, 18: 3, 19: 1, 20: 17, 21: 3, 22: 9, 23: 4, 24: 2, 25: 2, 26: 4, 27: 2, 28: 4, 29: 4, 30: 6, 31: 76, 32: 73, 33: 1, 34: 16, 35: 57, 36: 46, 37: 3, 38: 69, 39: 1, 40: 2, 41: 3, 42: 1, 43: 1, 44: 23}}

data = pd.DataFrame(d)
data = data[data['grad']!='Low Frequency']

rad = .2
conn_style = f'arc3,rad={rad}'

def offset(d, pos, dist = rad/2, loop_shift = .2):
    for (u,v),obj in d.items():
        if u!=v:
            par = dist*(pos[v] - pos[u])
            dx,dy = par[1],-par[0]
            x,y = obj.get_position()
            obj.set_position((x+dx,y+dy))
        else:
            x,y = obj.get_position()
            obj.set_position((x,y+loop_shift))

plt.figure(figsize = (20,10))
G = nx.from_pandas_edgelist(data, source='grad',
                                target='to', edge_attr='count',
                                create_using=nx.DiGraph())
weight = nx.get_edge_attributes(G, 'count')
pos = nx.shell_layout(G, scale=1)
nx.draw_networkx_nodes(G, pos, node_size=300, node_color='lightblue')
nx.draw_networkx_labels(G, pos=pos, font_color='red')
nx.draw_networkx_edges(G, pos=pos, edgelist=G.edges(), edge_color='black',
                       connectionstyle=conn_style)
d = nx.draw_networkx_edge_labels(G, pos=pos, edge_labels=weight)
offset(d,pos)
plt.gca().set_aspect('equal')

plt.show()

The result I get:

Here's what I get if I include the Low Frequency node and set rad to .1 instead of .2.

Here's an approach that doesn't change the aspect ratio:

import pandas as pd
import networkx as nx
import numpy as np
import matplotlib.pyplot as plt

d = {'grad': {0: 'CUHK', 1: 'CUHK', 2: 'CUHK', 3: 'CUHK', 4: 'CUHK', 5: 'CUHK', 6: 'CUHK', 7: 'CUHK', 8: 'CityU', 9: 'CityU', 10: 'CityU', 11: 'CityU', 12: 'CityU', 13: 'CityU', 14: 'CityU', 15: 'HKBU', 16: 'HKU', 17: 'HKU', 18: 'HKU', 19: 'HKU', 20: 'HKU', 21: 'HKU', 22: 'HKU', 23: 'HKUST', 24: 'HKUST', 25: 'HKUST', 26: 'HKUST', 27: 'HKUST', 28: 'HKUST', 29: 'HKUST', 30: 'HKUST', 31: 'Low Frequency', 32: 'Low Frequency', 33: 'Low Frequency', 34: 'Low Frequency', 35: 'Low Frequency', 36: 'Low Frequency', 37: 'Low Frequency', 38: 'Low Frequency', 39: 'PolyU', 40: 'PolyU', 41: 'PolyU', 42: 'PolyU', 43: 'PolyU', 44: 'PolyU'}, 'to': {0: 'CUHK', 1: 'CityU', 2: 'EduHK', 3: 'HKBU', 4: 'HKU', 5: 'HKUST', 6: 'LingU', 7: 'PolyU', 8: 'CityU', 9: 'EduHK', 10: 'HKBU', 11: 'HKU', 12: 'HKUST', 13: 'LingU', 14: 'PolyU', 15: 'HKBU', 16: 'CUHK', 17: 'CityU', 18: 'EduHK', 19: 'HKBU', 20: 'HKU', 21: 'HKUST', 22: 'PolyU', 23: 'CUHK', 24: 'CityU', 25: 'EduHK', 26: 'HKBU', 27: 'HKU', 28: 'HKUST', 29: 'LingU', 30: 'PolyU', 31: 'CUHK', 32: 'CityU', 33: 'EduHK', 34: 'HKBU', 35: 'HKU', 36: 'HKUST', 37: 'LingU', 38: 'PolyU', 39: 'CityU', 40: 'EduHK', 41: 'HKBU', 42: 'HKU', 43: 'LingU', 44: 'PolyU'}, 'count': {0: 13, 1: 6, 2: 3, 3: 6, 4: 5, 5: 3, 6: 2, 7: 6, 8: 4, 9: 1, 10: 5, 11: 2, 12: 1, 13: 2, 14: 7, 15: 2, 16: 2, 17: 4, 18: 3, 19: 1, 20: 17, 21: 3, 22: 9, 23: 4, 24: 2, 25: 2, 26: 4, 27: 2, 28: 4, 29: 4, 30: 6, 31: 76, 32: 73, 33: 1, 34: 16, 35: 57, 36: 46, 37: 3, 38: 69, 39: 1, 40: 2, 41: 3, 42: 1, 43: 1, 44: 23}}

data = pd.DataFrame(d)    

rad = .1
conn_style = f'arc3,rad={rad}'

def offset(d, pos, dist = rad/2, loop_shift = .2, asp = 1):
    for (u,v),obj in d.items():
        if u!=v:
            par = dist*(pos[v] - pos[u])
            dx,dy = par[1]*asp,-par[0]/asp
            x,y = obj.get_position()
            obj.set_position((x+dx,y+dy))
        else:
            x,y = obj.get_position()
            obj.set_position((x,y+loop_shift))

def sub(a,b):
    return a-b

def get_aspect(ax):
    # Total figure size
    figW, figH = ax.get_figure().get_size_inches()
    # Axis size on figure
    _, _, w, h = ax.get_position().bounds
    # Ratio of display units
    disp_ratio = (figH * h) / (figW * w)
    # Ratio of data units
    # Negative over negative because of the order of subtraction
    data_ratio = sub(*ax.get_ylim()) / sub(*ax.get_xlim())

    return disp_ratio / data_ratio

plt.figure(figsize = (20,10))
G = nx.from_pandas_edgelist(data, source='grad',
                                target='to', edge_attr='count',
                                create_using=nx.DiGraph())
weight = nx.get_edge_attributes(G, 'count')
pos = nx.shell_layout(G, scale=1)
nx.draw_networkx_nodes(G, pos, node_size=300, node_color='lightblue')
nx.draw_networkx_labels(G, pos=pos, font_color='red')
nx.draw_networkx_edges(G, pos=pos, edgelist=G.edges(), edge_color='black',
                       connectionstyle=conn_style)
d = nx.draw_networkx_edge_labels(G, pos=pos, edge_labels=weight)

offset(d, pos, asp = get_aspect(plt.gca()))

plt.show()

Resulting figure:

Thanks for the result, and this is a nice graph. Is the Aspect ratio & size of this figure be fixed? — kenny, Nov 30 '22 at 08:53
@kenny Without setting the aspect ratio as in `plt.gca().set_aspect('equal')`, it seems that the modified label placements that I produce with the `offset` function are incorrect. I'm not sure why that should be the case, and I haven't found a good way to make this work for the default aspect ratio. — Ben Grossmann, Nov 30 '22 at 12:38
@kenny You could also consider using the netgraph package, which is used in [this answer](https://stackoverflow.com/a/72184427/2476977) — Ben Grossmann, Nov 30 '22 at 12:41
@kenny Actually, I think I've made it work. See my latest edit. — Ben Grossmann, Nov 30 '22 at 13:08

How to label edges and avoid the edge overlapping in MultiDiGraph/DiGraph? (Networkx)

1 Answers1