Reordering and facetting for plotnine

[juanramonua]

Reordering labels according to their values by facet is complicated in plotnine. Now, as far as I know there is a trick that consists of creating an auxiliary index, creating the plot with the index and replacing each index with its correct label (e.g. scale_x_discrete(labels=...)).

In ggplot2 there is a more intuitive (ej. https://juliasilge.com/blog/reorder-within/) alternative combining the reorder_within(...) with scale_x_reordered() functions. My suggestion is to implement these functions in future versions to create this type of graphs in a more intuitive way.

Best regards and thanks.

[has2k1]

There is an internal reorder function that you can use inside aes.

from string import ascii_lowercase, ascii_uppercase
import numpy as np
import pandas as pd
from plotnine import *
np.random.seed(123)

n = 4
m = 10

df = pd.DataFrame({
    "x": list(ascii_lowercase[:m]) * n,
    "y": np.random.randint(1, 101, n * m),
    "g": np.repeat(list(ascii_uppercase[:n]), m)
})

sep = "::"

(ggplot(df, aes("reorder(x+sep+g, y)", "y", fill="x"))
 + geom_col()
 + labs(x="x")
 + facet_wrap("g", scales="free_x")
 + scale_x_discrete(labels=lambda lst: [l.split(sep)[0] for l in lst])
)

You can do some refactoring and use the internal function (not a good idea as the location might change in the future) to get the same result.

from string import ascii_lowercase, ascii_uppercase
import numpy as np
import pandas as pd
from plotnine import *
from plotnine.mapping.evaluation import reorder
np.random.seed(123)

n = 4
m = 10

df = pd.DataFrame({
    "x": list(ascii_lowercase[:m]) * n,
    "y": np.random.randint(1, 101, n * m),
    "g": np.repeat(list(ascii_uppercase[:n]), m)
})

sep = "::"

def reorder_within_facet(a, b, facet_var):
    _a = a.astype(str) + sep + facet_var.astype(str)
    res = reorder(_a, b)
    return res

def reordered_labels(labels):
    return [l.split(sep)[0] for l in labels]

(ggplot(df, aes("reorder_within_facet(x, y, g)", "y", fill="x"))
 + geom_col()
 + facet_wrap("g", scales="free_x")
 + scale_x_discrete(name="x", labels=reordered_labels)
)

The functions reorder_within_facet and scale_x_reordered cannot be included in plotnine because they are coupled. I think the best place for this is as a recipe in the documentation.