Some categorical plots fail when y is datetime

Categorical plots are not interchangeable when the dependent variable is np.datetime. Apart from stripplot and swarmplot they all may fail. Consider the following code:

import seaborn as sns
import numpy as np
dates = np.arange('2020-03', '2020-04', dtype='datetime64[D]')
cats = np.repeat("cat", len(dates))
for func in [sns.stripplot, sns.swarmplot, sns.boxplot, sns.boxenplot, sns.violinplot, sns.barplot]:
    try:
         func(cats, dates)
    except Exception as ex:
         print(str(func), type(ex), ex)

The above code results with the following lines printed:

<function boxplot at 0x7f520e7af598> <class 'numpy.core._exceptions.UFuncTypeError'> ufunc 'add' cannot use operands with types dtype('<M8[ns]') and dtype('<M8[ns]')
<function boxenplot at 0x7f520e7af730> <class 'numpy.core._exceptions.UFuncTypeError'> ufunc 'multiply' cannot use operands with types dtype('<M8[ns]') and dtype('float64')
<function violinplot at 0x7f520e7af620> <class 'TypeError'> invalid type promotion
<function barplot at 0x7f520e7af8c8> <class 'numpy.core._exceptions.UFuncTypeError'> ufunc 'add' cannot use operands with types dtype('<M8[ns]') and dtype('<M8[ns]')

The problem is that datetime variable is not identified as categorical and also not identified as non-numeric and therefore specific plotters attempt to perform operations that expect numeric input. Categorical plots docstrings currently don’t mention datetime y (only datetime x).

Probably the easiest fix to maintain consistency between all categorical plots would be changing the following method to identify dates as non-numeric (perhaps using pandas.api.types.is_numeric_dtype). https://github.com/mwaskom/seaborn/blob/562fd90216d5529ab8f26b388168e6ef260d7e6c/seaborn/categorical.py#L330-L335 However this will also disable stripplot and swarmplot with dates.

[isomorpheme]

I think I’m having a similar issue: I have a dataframe with a country categorical column, and a created_at datetime column. Using catplot(data, kind='swarm', x='country', y='created_at') works fine, but when I use kind='violin', Seaborn complains with the aforementioned TypeError: Neither the xnory variable appears to be numeric. I would expect otherwise, because I can still e.g. plot a distribution for all dates using distplot, and a violin plot does essentially that, just per category.

[MaozGelbart]

I’ll close this as the reason for failures is now clear for the user. Will open a separate issue with an example regarding qualitative colormap when hue is datetime. Thanks!