Fix setting edge colors for Chord graphs using an explicit mapping

[schumann-tim]

and normalization (which is turned on by default).

Details:
The color map's vmin/vmax was computed based on factors of the whole range
whereas the palette only used the factors from the actually used values.

When using color normalization this lead to:
-- LineCollection normalizing via vmin/vmax computed above, reflecting the
whole range.
-- ListedColormap is transforming the normalized color values provided by the
LineCollection using its own size (N). However that size N was based on the
actually used values and hence too small.

[philippjfr]

Thanks for the fix, it's very much appreciated.

[philippjfr]

Presumably for the bokeh backend this was already handled correctly?

[schumann-tim]

Presumably for the bokeh backend this was already handled correctly?
I didn't try it out with bokeh; when I installed holoviews with conda, apparently a too-old version of bokeh was bundled. As I primarily wanted to generate the graphs as svgs I didn't follow up. I can check though.

Would a minimal example be useful for regression testing? (not sure how the current set-up of the project is).

[philippjfr]

Would a minimal example be useful for regression testing?

Definitely, you can look at the existing tests in holoviews/tests/plotting/matplotlib/testgraphplot.py TestMplChordPlot for some examples.

[schumann-tim]

To keep the conversation going, here's a repro for the bug.
I'm unfortunately traveling without access to my computer. Will pick this up when i'm back in ~2 weeks.

import os
import pandas as pd
import holoviews as hv
from holoviews import opts, dim

hv.extension('matplotlib')

edges = [{'source': 0, 'target': 2, 'value': 168},
{'source': 0, 'target': 3, 'value': 150},
{'source': 1, 'target': 3, 'value': 50},
{'source': 1, 'target': 2, 'value': 100}]

nodes = [{'index': 0, 'group': 'clonal_id', 'name': 'grey'},
{'index': 1, 'group': 'clonal_id', 'name': 'red'},
{'index': 2, 'group': 'hits', 'name': 'blue'},
{'index': 3, 'group': 'hits', 'name': 'green'}]

Edges reference by the source's node-id -- so include them as well.

color_map = {'grey':'grey', '0': 'grey',
'red': 'red', '1': 'red',
'blue': 'blue', '2': 'blue',
'green': 'green', '3': 'green'}

links_df = pd.DataFrame(edges)
nodes_ds = hv.Dataset(pd.DataFrame(nodes), 'index')
graph = hv.Chord((links_df, nodes_ds), kdims=['source', 'target']).opts(
opts.Chord(
cmap=color_map, edge_cmap=color_map,
edge_color=(dim('source')).astype(str), label_index='name',
node_color=dim('name').str(), normalize=True))

output_path = '/tmp/repro.pdf'
hv.save(graph, output_path)

[philippjfr]

Just going to merge.