Improve default node padding on Sankey

Generating a Sankey diagram with many nodes results in a plot with many overlaps of nodes and crisscrossing of edges. It’s nearly impossible to interpret the data with so many overlapping nodes and crisscrossing edges.

Example:

sankey = hv.Sankey(data, kdims=['source', 'target'], vdims='value') \
               .opts(width=1000, height=700, label_position='left', edge_color='target', node_color='index', cmap='tab20')

Produces the following plot:

Where data looks like:

source	target	type	value
B45	C0	2	339
B45	C6	2	318
B45	C78	2	185
B45	C52	2	179
B45	C67	2	158
B45	C92	2	144
B60	C71	2	131
B45	C65	2	122
B45	C13	2	121
B49	C71	2	107
B45	C93	2	86
B45	C9	2	83
B14	C92	2	82
B60	C69	2	82
B78	C93	2	81
B30	C91	2	75
B92	C9	2	74
B45	C83	2	73
B49	C41	2	73
B80	C45	2	73
B45	C71	2	70
B38	C45	2	65
B13	C71	2	62
B45	C60	2	61
B60	C93	2	55
B60	C99	2	55
B49	C33	2	53
B45	C34	2	52
B83	C56	2	52
B92	C6	2	51
…	…	…	…
A5	B77	1	48
A5	B17	1	45
A5	B9	1	40
A5	B37	1	38
A5	B68	1	38
A5	B84	1	35
A5	B56	1	31
A5	B59	1	31
A5	B74	1	30
A5	B25	1	28
A5	B11	1	21
A5	B34	1	20
A5	B61	1	20
A5	B98	1	19
A5	B12	1	15
A5	B6	1	15
A5	B33	1	14
A5	B19	1	13
A5	B23	1	9
A5	B46	1	9
A5	B44	1	7
A5	B36	1	6
A5	B70	1	6
A5	B42	1	5
A5	B53	1	3
A5	B26	1	2
A5	B3	1	2
A5	B8	1	2
A5	B15	1	1
A5	B16	1	1

997 rows × 4 columns

[odoublewen]

I also had this question. Seems like the code example in the Holoviews docs also offers an example of non-optimal behavior, in that nodes overlap unnecessarily with edges. I would expect the method to try to minimize not just node-node overlap but also node-edge overlap.

For example, why is the “Early Career Researcher” node superimposed on top of the edge that connects “PhD” and “Career Outside Science”?

Here’s an even simpler example:

nodes = ["node0","node1","node2", "node3", "node4"]
nodes = hv.Dataset(enumerate(nodes), 'index', 'label')
edges = [(0,1, 100),
         (1,2, 90),
        (1,3, 10),
        (3,4,10)]
reads = hv.Sankey((edges, nodes), ['From', 'To'])
reads.opts(opts.Sankey(labels='label', node_padding=10, label_position='right', width=900, height=300, cmap='Set1',
            edge_color=dim('To').str(), node_color=dim('index').str()))

Why does node3 have to be plotted inside of the edge connecting node1 and node2?

(edit: to fixed the node labels in my example)

[philippjfr]

For example here is some synthetic data

node_padding=10

node_padding=2