Is there a technique to create columns headers for a Plotly Sankey Diagram, similar to Tableau

Question

Is there a technique to plot a time series flow diagram with the column nodes representing the start date for each month, the values representing the counts for each type, and labels representing the types (i.e. Consumer, Home Office, Corporate, Small Business in the example image below)?

Plotly has a few examples on how to create a Sankey Diagram in Python. Adding dates as the column headers, similar to the Tableau example, Superstore Interactive Demo, would provide more clarity to the Sankey diagram. For example, "Level 0 Region" would be replaced with "Jan. 1, 2022" and "Level 2 Customer Segment" would be replaced with "Feb. 1, 2022."

Answer 1

• this is an enhancement to Plotly.py Sankey Diagrams - Controlling Node Destination
• your stated requirement is to create columns by date, use date part of concatenated sankey node for column
• clearly formatting can be further beautified. This shows how columns can be defined and annotated

sample data

from_date	to_date	from_type	to_type	value	source	target
2022-01-01 00:00:00	2022-02-01 00:00:00	Consumer	Home Office	3	Consumer_20220101	Home Office_20220201
2022-01-01 00:00:00	2022-03-01 00:00:00	Consumer	Corporate	6	Consumer_20220101	Corporate_20220301
2022-01-01 00:00:00	2022-03-01 00:00:00	Small Business	Corporate	21	Small Business_20220101	Corporate_20220301
2022-01-01 00:00:00	2022-04-01 00:00:00	Consumer	Home Office	14	Consumer_20220101	Home Office_20220401
2022-02-01 00:00:00	2022-03-01 00:00:00	Corporate	Consumer	20	Corporate_20220201	Consumer_20220301

solution

import pandas as pd
import numpy as np
import plotly.graph_objects as go

ms = pd.date_range("1-jan-2022", freq="MS", periods=4)
types = ["Consumer", "Home Office", "Corporate", "Small Business"]

# simulate some data, date and type to date and type
s = 50
df = pd.DataFrame(
    {
        "from_date": np.random.choice(ms, s),
        "to_date": np.random.choice(ms, s),
        "from_type": np.random.choice(types, s),
        "to_type": np.random.choice(types, s),
        "value": np.random.randint(1, 20, s),
    }
).loc[
    # remove invalid combis from random generation
    lambda d: (d["to_date"] > d["from_date"]) & (d["from_type"] != d["to_type"])
].groupby(
    ["from_date", "to_date", "from_type", "to_type"], as_index=False
).sum()

# start of solution, define source and target of sankey from column concat
df = df.assign(source=lambda d: d["from_type"] + "_" + d["from_date"].dt.strftime("%Y%m%d"),
          target=lambda d: d["to_type"] + "_" + d["to_date"].dt.strftime("%Y%m%d"),
         )


def factorize(s):
    a = pd.factorize(s, sort=True)[0]
    return (a + 0.01) / (max(a) + 0.1)


# unique nodes
nodes = np.unique(df[["source", "target"]], axis=None)
nodes = pd.Series(index=nodes, data=range(len(nodes)))
# work out positioning of nodes
nodes = (
    nodes.to_frame("id")
    .assign(
        y=lambda d: factorize(d.index.to_series().apply(lambda s: s.split("_")[0])),
        x=lambda d: factorize(d.index.to_series().apply(lambda s: s.split("_")[1])),
    )
)

# now simple job of building sankey
fig = go.Figure(
    go.Sankey(
        arrangement="snap",
        node={"label": nodes.index.to_series().apply(lambda s: s.split("_")[0]), "x": nodes["x"], "y": nodes["y"]},
        link={
            "source": nodes.loc[df["source"], "id"],
            "target": nodes.loc[df["target"], "id"],
            "value": df["value"],
        },
    )
)

for i, x in nodes["x"].drop_duplicates().iteritems():
    fig.add_annotation(x=x, y=1.4, text=i.split("_")[1], showarrow=False)
    
fig