Demo: Externally Defined Layouts

This notebook computes a layout using NetworkX and sends it to Graphistry for subsequent interactive analysis.



In [1]:

    
import graphistry
#graphistry.register(key='MY_KEY', server='labs.graphistry.com') #https://www.graphistry.com/api-request
import networkx as nx
import pandas as pd

Make a NetworkX Graph



In [2]:

    
num_nodes = 1000
G=nx.Graph()
G.add_nodes_from(range(num_nodes))
Edges = [(x, (x * 2) % (num_nodes/5)) for x in range(num_nodes)] + [(x, (x + 1) % 20) for x in range(20)]
G.add_edges_from(Edges)
G









    Out[2]:





<networkx.classes.graph.Graph at 0x10de7e150>

Use a NetworkX layout



In [3]:

    
%time
pos=nx.fruchterman_reingold_layout(G)
pos[0]









    



CPU times: user 4 µs, sys: 1 µs, total: 5 µs
Wall time: 9.06 µs






    Out[3]:





array([0.3766924 , 0.36150154], dtype=float32)

Combine into node, edge dataframes



In [4]:

    
def pos2df (pos):
    nodes = pd.DataFrame({'key': pos.keys(), 'pos_0': [pos[k][0] for k in pos.keys()], 'pos_1': [pos[k][1] for k in pos.keys()]})
    nodes.columns = ['key', 'x', 'y']
    return nodes



In [5]:

    
nodes = pos2df(pos)
nodes[:3]



In [6]:

    
edges = pd.DataFrame(Edges)
edges.columns = ['src', 'dst']
edges[:3]

Autolayout mode



In [7]:

    
g = graphistry.nodes(nodes).edges(edges).bind(source='src', destination='dst', node='key')
g.plot()









    Out[7]:

Predefined layout mode

As the node table provides "x" and "y" columns, they will be automatically used as starting positions. To prevent the automatic layout algorithm from moving the nodes on load, we also set the URL parameter "play" to 0 seconds. (Both settings will likely change.)



In [8]:

    
g2 = g.settings(url_params={'play':0})
g.plot()









    Out[8]:

For fun, here's a circular layout



In [9]:

    
pos2=nx.circular_layout(G, scale=100)
nodes2 = pos2df(pos2)
g2.nodes(nodes2).plot()









    Out[9]:



In [ ]:

	key	x	y
0	0	0.376692	0.361502
1	1	0.483908	0.427637
2	2	0.588375	0.491546