catplot with numeric hue and hue_order: empty legend handles

[jhncls]

Tested with Seaborn 0.13.2, pandas 2.2.1

When the hue values are numeric, hue_order isn't respected in the plot. The legend does respect the order, but with empty legend handles.

import seaborn as sns
import pandas as pd

tips = sns.load_dataset('tips')
sns.catplot(tips, kind='box', x='time', y='tip', hue='size', hue_order=[2, 3, 4])

Making the hue column of type pd.Categorical, but still numeric, does respect the hue order. But again with empty legend handles. The default palette changes to categorical.

import seaborn as sns
import pandas as pd

tips = sns.load_dataset('tips')
tips['size'] = pd.Categorical(tips['size'])
sns.catplot(tips, kind='box', x='time', y='tip', hue='size', hue_order=[2, 3, 4])

[mwaskom]

When the hue values are numeric, hue_order isn't respected in the plot.

This is expected

The legend does respect the order, but with empty legend handles.

This is probably an artifact of the different handling for the catplot legend

[jhncls]

The numeric hue values behave a bit unexpected when somebody wanted to superimpose a stripplot on a boxplot via a catplot. The plots don't align if not all hue values are present in all subplots. It only aligns well for non-numeric hue values of type pd.Categorical.

import seaborn as sns
tips = sns.load_dataset('tips')
common_kws = dict(x='time', y='tip', hue='size', dodge=True, palette='turbo')
g = sns.catplot(tips, kind='box', col='smoker', fill=False, **common_kws)
for s, ax in g.axes_dict.items():
    sns.stripplot(tips[tips['smoker'] == s], s=8, legend=False, ax=ax, **common_kws)

Anyway, it is a bit surprising that hue_order can be used for subsets, supersets and ordering, but only for strings. (I am not asking to make any changes. I understand it is complicated, where people sometimes interpret numbers as discrete and sometimes as continuous.)