Python Chord Diagram in Python/v3

How to make an interactive chord diagram in Python with Plotly and iGraph.


Note: this page is part of the documentation for version 3 of Plotly.py, which is not the most recent version.
See our Version 4 Migration Guide for information about how to upgrade.

Chord Diagrams with Plotly

A circular layout places the graph nodes uniformly on a circle. In this example we illustrate how to draw the graph edges in order to avoid a cluttered visualization.

Eurovision Song Contest Example

As an example, we consider a circular graph with the European countries as nodes. Among these countries, some qualified for the grand final Eurovision Song Contest.

Each european country is a jury member and rates some contestants on a scale from 1 to 12 (in 2015 a contestant from Australia led to adding this country to the graph).

There is a directed edge from a jury member country to a contestant country if the contestant acquired at least one point from the jury country voters.

The jury member countries are placed uniformly, in alphabetical order, on the unit circle. If there is an edge between two nodes, then we draw a cubic Bézier curve having as the first and the last control point the given nodes.

To avoid cluttered edges we adopted the following procedure in choosing the interior control points for the Bézier curve:

  • we consider five equally spaced points on the unit circle, corresponding to the angles $0, \:\pi/4$ $\pi/2,\: 3\pi/4, \:\pi$: $$P_1(1,0), \: P_2=(\sqrt{2}/2,\: \sqrt{2}/2),\: P_3(0,1), \: P_4=(-\sqrt{2}/2, \sqrt{2}/2),\: P_5(-1,0)$$

  • define a list, Dist, having as elements the distances between the following pairs of points: $$(P_1, P_1), \:(P_1, P_2), \: (P_1, P_3),\: (P_1, P_4),\: (P_1, P_5)$$

  • In order to assign the control poligon to the Bézier curve that will be the edge between two connected nodes, V[i], V[j], we compute the distance between these nodes, and deduce the interval $k$, of two consecutive values in Dist, this distance belongs to.

  • Since there are four such intervals indexed $k=0,1,2,3$, we define the control poligon as follows: $${\bf b}_0=V[i],\:\: {\bf b}_1=V[i]/param,\:\: {\bf b}_2=V[j]/param, \:\:{\bf b}_3=V[j],$$ where param is chosen from the list: params=[1.2, 1.5, 1.8, 2.1].

Namely, if the distance(V[i], V[j]), belongs to the $K^{th}$ interval associated to Dist, then we choose param= params[K].

We processed data provided by Eurovision Song Contest, and saved the corresponding graph in a gml file. Now we can read the gml file and define an igraph.Graph object.

Install the Python libraries with sudo pip install python-igraph and sudo pip install networkx.

In [1]:
import igraph as ig

G = ig.Graph.Read_GML('Eurovision15.gml')

Define the list of nodes (vs stands for vertices):

In [2]:
V=list(G.vs)
G.vs.attributes()# list node attributes
Out[2]:
['id', 'label']

Define the label list. Labels will be displayed in the Plotly plot:

In [3]:
labels=[v['label']  for v in V]

G.es is the sequence of graph edges

In [4]:
G.es.attributes()# the edge attributes
Out[4]:
['weight']

Get the edge list as a list of tuples, having as elements the end nodes indices:

In [5]:
E=[e.tuple for e in G.es]# list of edges
len(E)
Out[5]:
400

Get the list of Contestant countries:

In [6]:
ContestantLst=[G.vs[e[1]] for e in E]
Contestant=list(set([v['label'] for  v in ContestantLst]))
len(Contestant)
Out[6]:
25

Get the node positions, assigned by the circular layout:

In [7]:
layt=G.layout('circular') #circular layout

layt is a list of 2-elements lists, representing the coordinates of nodes placed on the unit circle:

In [8]:
L=len(layt)
layt[7]
Out[8]:
[0.45399049973954686, 0.8910065241883678]

Define the list of edge weights:

In [9]:
Weights= map(int, G.es["weight"])

In the sequel we define a few functions that lead to the edge definition as a Bézier curve:
dist(A,B) computes the distance between two 2D points, A, B:

In [10]:
import numpy as np

def dist (A,B):
    return np.linalg.norm(np.array(A)-np.array(B))
dist(layt[0], layt[5])
Out[10]:
0.76536686473017945

Define the list Dist of threshold distances between nodes on the unit circle and the list of parameters for interior control points:

In [11]:
Dist=[0, dist([1,0], 2*[np.sqrt(2)/2]), np.sqrt(2),
      dist([1,0],  [-np.sqrt(2)/2, np.sqrt(2)/2]), 2.0]
params=[1.2, 1.5, 1.8, 2.1]

The function get_idx_interv returns the index of the interval the distance d belongs to:

In [12]:
def get_idx_interv(d, D):
    k=0
    while(d>D[k]):
        k+=1
    return  k-1

Below are defined the function deCasteljau and BezierCv. The former returns the point corresponding to the parameter t, on a Bézier curve of control points given in the list b.

The latter returns an array of shape (nr, 2) containing the coordinates of nr points evaluated on the Bézier curve, at equally spaced parameters in [0,1].

For our purpose the default number of points evaluated on a Bézier edge is 5. Then setting the Plotly shape of the edge line as spline, the five points are interpolated.

In [13]:
class InvalidInputError(Exception):
    pass

def deCasteljau(b,t):
    N=len(b)
    if(N<2):
        raise InvalidInputError("The  control polygon must have at least two points")
    a=np.copy(b) #shallow copy of the list of control points 
    for r in range(1,N):
        a[:N-r,:]=(1-t)*a[:N-r,:]+t*a[1:N-r+1,:]
    return a[0,:]

def BezierCv(b, nr=5):
    t=np.linspace(0, 1, nr)
    return np.array([deCasteljau(b, t[k]) for k in range(nr)])

Finally we set data and layout for the Plotly plot of the circular graph:

In [14]:
import plotly.plotly as py
import plotly.graph_objs as go

node_color=['rgba(0,51,181, 0.85)'  if v['label'] in Contestant else '#CCCCCC' for v in G.vs]
line_color=['#FFFFFF'  if v['label'] in Contestant else 'rgb(150,150,150)' for v in G.vs]
edge_colors=['#d4daff','#84a9dd', '#5588c8', '#6d8acf']

Define the lists of x, respectively y-coordinates of the nodes:

In [15]:
Xn=[layt[k][0] for k in range(L)]
Yn=[layt[k][1] for k in range(L)]

On each Bézier edge, at the point corresponding to the parameter $t=0.9$, one displays the source and the target node labels, as well as the number of points (votes) assigned by source to target.

In [16]:
lines=[]# the list of dicts defining   edge  Plotly attributes
edge_info=[]# the list of points on edges where  the information is placed

for j, e in enumerate(E):
    A=np.array(layt[e[0]])
    B=np.array(layt[e[1]])
    d=dist(A, B)
    K=get_idx_interv(d, Dist)
    b=[A, A/params[K], B/params[K], B]
    color=edge_colors[K]
    pts=BezierCv(b, nr=5)
    text=V[e[0]]['label']+' to '+V[e[1]]['label']+' '+str(Weights[j])+' pts'
    mark=deCasteljau(b,0.9)
    edge_info.append(go.Scatter(x=[mark[0]],
                             y=[mark[1]],
                             mode='markers',
                             marker=dict( size=0.5,  color=edge_colors),
                             text=text,
                             hoverinfo='text'
                             )
                    )
    lines.append(go.Scatter(x=pts[:,0],
                         y=pts[:,1],
                         mode='lines',
                         line=dict(color=color,
                                  shape='spline',
                                  width=Weights[j]/5#The  width is proportional to the edge weight
                                 ),
                        hoverinfo='none'
                       )
                )

trace2=go.Scatter(x=Xn,
           y=Yn,
           mode='markers',
           name='',
           marker=dict(symbol='circle',
                         size=15,
                         color=node_color,
                         line=dict(color=line_color, width=0.5)
                         ),
           text=labels,
           hoverinfo='text',
           )

axis=dict(showline=False, # hide axis line, grid, ticklabels and  title
          zeroline=False,
          showgrid=False,
          showticklabels=False,
          title=''
          )

def make_annotation(anno_text, y_coord):
    return dict(showarrow=False,
                      text=anno_text,
                      xref='paper',
                      yref='paper',
                      x=0,
                      y=y_coord,
                      xanchor='left',
                      yanchor='bottom',
                      font=dict(size=12)
                     )

anno_text1='Blue nodes mark the countries that are both contestants and jury members'
anno_text2='Grey nodes mark the countries that are only jury members'
anno_text3='There is an edge from a Jury country to a contestant country '+\
           'if the jury country assigned at least one vote to that contestant'
width=800
height=850
title="A circular graph associated to Eurovision Song Contest, 2015<br>Data source:"+\
"<a href='http://www.eurovision.tv/page/history/by-year/contest?event=2083#Scoreboard'> [1]</a>"
layout=go.Layout(title= title,
              font= dict(size=12),
              showlegend=False,
              autosize=False,
              width=width,
              height=height,
              xaxis=dict(axis),
              yaxis=dict(axis),
              margin=dict(l=40,
                            r=40,
                            b=85,
                            t=100,
                          ),
              hovermode='closest',
              annotations=list([make_annotation(anno_text1, -0.07),
                                       make_annotation(anno_text2, -0.09),
                                       make_annotation(anno_text3, -0.11)]
                                     )
              )

data=lines+edge_info+[trace2]
fig=go.Figure(data=data, layout=layout)
py.iplot(fig, filename='Eurovision-15')
Out[16]:

Now we can compare to the Eurovision graph as a networkx.Graph:

In [17]:
import networkx as nx
import matplotlib.pyplot as plt
%matplotlib inline

G=nx.DiGraph()
G.add_nodes_from(range(L))
G.add_edges_from(E)

plt.figure(figsize=(18,18))
nx.draw_circular(G,node_color='g', edge_color='#909090', node_size=900)
plt.axis('equal')
Out[17]:
(-1.5, 1.5, -1.5, 1.5)