How to compute percentage per variable ?

[gigaturbo]

Hi guys,

I'm ploting a bar graph in fill mode, I have a main category and I'm counting occurences of a second category.
When I plot and add the labels, I end up with percentages relative ti the whole datasetn and not relative to category 1 (as I was expecting).

Here's the result:

I used a fill mapping for category2 and face_wrap for category 1.

I would like that in each category, percentages sum up to 100% (and bar height are also wrong I think in my actual setup).

How to do this?

[has2k1]

Please provide a minimal code snippet.

[gigaturbo]

!pip install plotnine
import pandas as pd
import plotnine as p9

d = pd.DataFrame({
    "cat1": ["A","A","B","A","A","B","B","A","B","B","B","A","B","B","B","A","B"],
    "cat2": ["W","X","Y","Z","W","X","Y","Z","W","X","Y","Z","W","X","Y","Z","X"]
}).assign(x=1)

g = (
        p9.ggplot(d)
        + p9.geom_bar(p9.aes(x="x", fill="cat2"), stat="count", position="fill")
        + p9.geom_text(
            p9.aes(
                x="x",
                fill="cat2",
                label=p9.after_stat("count/sum(count)*100")
            ),
            stat="count",
            position=p9.position_fill(vjust=0.5),
            format_string="{:.0f}%",
        )
        + p9.facet_wrap("cat1")
        + p9.scale_y_continuous(labels=lambda lbs: [f"{int(v * 100):d}%" for v in lbs])
    )

g

I got this, where percentage sum up to 100% on bot A+B categories:

But I want a 100% sum for A and B separately.

[has2k1]

When grouping gets complicated, like here where there are nested groups, it is better (and it may be the only way) to precompute the values and have them in a dataframe as well so that a straight mapping to that column is enough.

However in this case, you can get away with this solution that relies on an internal variable PANEL.

import pandas as pd
import plotnine as p9
from mizani.labels import label_percent

d = pd.DataFrame({
    "cat1": ["A","A","B","A","A","B","B","A","B","B","B","A","B","B","B","A","B"],
    "cat2": ["W","X","Y","Z","W","X","Y","Z","W","X","Y","Z","W","X","Y","Z","X"]
}).assign(x=1)

def sum_by(col: pd.Series, by: pd.Series):
    df = pd.DataFrame({col.name: col, "by": by})
    return df.groupby("by", observed=False)[col.name].transform("sum")

(
    p9.ggplot(d)
    + p9.geom_bar(p9.aes(x="x", fill="cat2"), stat="count", position="fill")
    + p9.geom_text(
        p9.aes(
            x="x",
            group="cat2",
            label=p9.after_stat("100*count/sum_by(count, PANEL)")
        ),
        stat="count",
        position=p9.position_fill(vjust=0.5),
        format_string="{:.0f}%",
    )
    + p9.facet_wrap("cat1")
    + p9.scale_y_continuous(labels=label_percent())
)

[gigaturbo]

Thanks for the answer, I did not knew about the internal variables!

[has2k1]

It is not part of the public API, so you should really rely on it.