stacked + grouped bar chart

Question

I'm trying to create a bar chart using plotly in python, which is both stacked and grouped.
Toy example (money spent and earned in different years):

import pandas as pd
import plotly.graph_objs as go

data = pd.DataFrame(
    dict(
        year=[2000,2010,2020],
        var1=[10,20,15],
        var2=[12,8,18],
        var3=[10,17,13],
        var4=[12,11,20],
    )
)

fig = go.Figure(
    data = [
        go.Bar(x=data['year'], y=data['var1'], offsetgroup=0, name='spent on fruit'),
        go.Bar(x=data['year'], y=data['var2'], offsetgroup=0, base=data['var1'], name='spent on toys'),
        go.Bar(x=data['year'], y=data['var3'], offsetgroup=1, name='earned from stocks'),
        go.Bar(x=data['year'], y=data['var4'], offsetgroup=1, base=data['var3'], name='earned from gambling'),
    ]
)
fig.show()   

The result seems fine at first: But watch what happens when I turn off e.g. "spent on fruit": The "spent on toys" trace remains floating instead of starting from 0.
Can this be fixed? or maybe the whole offsetgroup + base approach won't work here. But what else can I do?
Thanks!

Update: according to this Github issue, stacked, grouped bar plots are being developed for future plotly versions, so this probably won't be an issue anymore.

Answer 1

Plotly Express (part of recent plotly library version) offers a facet_col parameter for its bar chart (and other charts as well), which allows one to set an additional grouping column:

Values from this column or array_like are used to assign marks to facetted subplots in the horizontal direction.

To make it work I had to reshape the example data:

import pandas as pd

data = pd.DataFrame(
    dict(
        year=[*[2000, 2010, 2020]*4],
        var=[*[10, 20, 15], *[12, 8, 18], *[10, 17, 13], *[12, 11, 20]],
        names=[
            *["spent on fruit"]*3,
            *["spent on toys"]*3,
            *["earned from stocks"]*3,
            *["earned from gambling"]*3,
        ],
        groups=[*["subgroup1"]*6, *["subgroup2"]*6]
    )
)

	year	var	names	groups
0	2000	10	spent on fruit	subgroup1
1	2010	20	spent on fruit	subgroup1
2	2020	15	spent on fruit	subgroup1
3	2000	12	spent on toys	subgroup1
4	2010	8	spent on toys	subgroup1
5	2020	18	spent on toys	subgroup1
6	2000	10	earned from stocks	subgroup2
7	2010	17	earned from stocks	subgroup2
8	2020	13	earned from stocks	subgroup2
9	2000	12	earned from gambling	subgroup2
10	2010	11	earned from gambling	subgroup2
11	2020	20	earned from gambling	subgroup2

Once it's in this format (I believe this is called the "tall format") you can plot it with one function call:

import plotly_express as px

fig = px.bar(data, x="groups", y="var", facet_col="year", color="names")
fig.show()

If you want to hide the subgroup labels you can update the x-axis:

fig.update_xaxes(visible=False)

Answer 2

There doesn't seem to be a way to create both stacked and grouped bar charts in Plotly, but there is a workaround that might resolve your issue. You will need to create subgroups, then use a stacked bar in Plotly to plot the bars one at a time, plotting var1 and var2 with subgroup1, and var3 and var4 with subgroup2.

This solution gives you the functionality you want, but changes the formatting and aesthetic of the bar chart. There will be equal spacing between each bar as from Plotly's point of view these are stacked bars (and not grouped bars), and I couldn't figure out a way to eliminate the subgroup1 and subgroup2 text without also getting rid of the years in the x-axis ticks. Any Plotly experts please feel free to chime in and improve my answer!

import pandas as pd
import plotly.graph_objs as go

df = pd.DataFrame(
    dict(
        year=[2000,2010,2020],
        var1=[10,20,15],
        var2=[12,8,18],
        var3=[10,17,13],
        var4=[12,11,20],
    )
)
        
fig = go.Figure()

fig.update_layout(
    template="simple_white",
    xaxis=dict(title_text="Year"),
    yaxis=dict(title_text="Count"),
    barmode="stack",
)

groups = ['var1','var2','var3','var4']
colors = ["blue","red","green","purple"]
names = ['spent on fruit','spent on toys','earned from stocks','earned from gambling']

i = 0
for r, n, c in zip(groups, names, colors):
    ## put var1 and var2 together on the first subgrouped bar
    if i <= 1:
        fig.add_trace(
            go.Bar(x=[df.year, ['subgroup1']*len(df.year)], y=df[r], name=n, marker_color=c),
        )
    ## put var3 and var4 together on the first subgrouped bar
    else:
        fig.add_trace(
            go.Bar(x=[df.year, ['subgroup2']*len(df.year)], y=df[r], name=n, marker_color=c),
        )
    i+=1

fig.show()