I'm trying to create a directed graph using networkx and bokeh, however, I also want to show the arrows for each out-edge. I found that the Holoviews library has the ability to add a 'directed=true' parameter to its graph constructor. However, I also want to utilize Bokeh's design features such as adjusting node color/size based on previously set node-attributes. The latter only works if I use Bokeh's from_networkx() to get a bokeh.models.renderers.GraphRender object, and then use its attributes node_renderer and edge_renderer.
The issue is when using Holoviews' renderer to specify Bokeh as the backend, it returns a bokeh.plotting.figure.Figure instead of the GraphRenderer. Ultimately, I want to be know how to be able to control the node size/color based on some attributes possible through Bokeh, and simultaneously use Holoviews to display the arrowheads on each edge.
import networkx as nx
import holoviews as hv
from holoviews import opts
hv.extension('bokeh')
from bokeh.io import show, output_file
# ... some code for I/O ...
G = nx.DiGraph(edgeList) # Directed networkx graph
# Set Node/Edge attributes to display upon hover
numConnections = {k:v for k,v in G.out_degree()}
nx.set_node_attributes(G, numConnections, name='numConnections')
# Returns Holoviews graph
hvGraph = hv.Graph.from_networkx(G, nx.spring_layout).opts(tools=['hover'], directed=True, arrowhead_length=0.01)
# Renders Holoviews graph into bokeh.plotting.figure.Figure
hvRendered = hv.render(hvGraph, 'bokeh')
output_file("out.html")
show(hvRendered)
# # The below code runs as expected using Bokeh only, and not Holoviews
# # to produce the directed graph (without arrowed edges):
# from bokeh.models import Plot, Range1d, MultiLine, Circle
# from bokeh.models import LinearColorMapper, ColorBar, BasicTicker
# import bokeh.models as bm
# from bokeh.models.graphs import from_networkx
# from bokeh.models.graphs import NodesAndLinkedEdges, EdgesAndLinkedNodes
# # Returns GraphRenderer from bokeh.models.renderers.DateRenderer
# graphRenderer = from_networkx(G, nx.spring_layout)
# mapper = LinearColorMapper(palette="Viridis256", low=76, high=0)
# # Node size/color when unselected / selected / hover
# graphRenderer.node_renderer.glyph = Circle(
# size='node_size',
# fill_color= {'field': "numConnections", "transform": mapper},
# fill_alpha=.8
# )
# graphRenderer.node_renderer.selection_glyph = Circle(
# size=25,
# fill_color=Inferno6[4]
# )
# graphRenderer.node_renderer.hover_glyph = Circle(
# size=20,
# fill_color=Inferno6[3]
# )
# # Edge size/color when unselected / selected / hover
# graphRenderer.edge_renderer.glyph = MultiLine(
# line_color="#CCCCCC",
# line_alpha=0.8,
# line_width=3
# )
# graphRenderer.edge_renderer.selection_glyph = MultiLine(
# line_color=Inferno6[4],
# line_width=4
# )
# graphRenderer.edge_renderer.hover_glyph = MultiLine(
# line_color=Inferno6[3],
# line_width=4
# )
# graphRenderer.node_renderer.data_source.data['numConnections'] = [v for k,v in
nx.get_node_attributes(G,'numConnections').items()]
# graphRenderer.selection_policy = NodesAndLinkedEdges()
# graphRenderer.inspection_policy = NodesAndLinkedEdges()
# bar = ColorBar(color_mapper=mapper, location=(0,0), title='#connections')
# # Create Bokeh Plot
# plot = Plot(
# plot_width=20,
# plot_height=20,
# x_range=Range1d(-1.1,1.1),
# y_range=Range1d(-1.1,1.1)
# )
# plot.add_tools(
# bm.HoverTool(tooltips=[("#connections", "@numConnections")]),
# bm.TapTool(),
# bm.BoxSelectTool()
# )
# plot.renderers.append(graphRenderer)
# output_file("bokeh.html")
# show(plot)
After rendering the Holoview graph into a Bokeh Figure (not a models.GraphRenderer), if I try to call the attribute node_renderer using the rendered Bokeh Figure object, it obviously throws an exception.
Traceback (most recent call last): File "holoview.py", line 106, in hvRenderedGraph.node_renderer.selection_glyph = Circle() AttributeError: 'BokehRenderer' object has no attribute 'node_renderer'
