Big-endian issue with pandas 2.1.0

[ManonMarchand]

Hello Seaborn devs and thanks for your software!

There is an issue with plotting and pandas 2.1.0 :

Doing this

import numpy as np
import seaborn as sns
x = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
y = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
sns.lineplot(x=x, y=y)

Returns the following error

Unroll for traceback

ValueError                                Traceback (most recent call last)
Cell In[3], line 5
      3 x = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
      4 y = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
----> 5 sns.lineplot(x=x, y=y)

File [~/.local/lib/python3.10/site-packages/seaborn/relational.py:638](https://file+.vscode-resource.vscode-cdn.net/home/manon.marchand/Documents/cds-astro-github/tutorials/Notebooks/~/.local/lib/python3.10/site-packages/seaborn/relational.py:638), in lineplot(data, x, y, hue, size, style, units, palette, hue_order, hue_norm, sizes, size_order, size_norm, dashes, markers, style_order, estimator, errorbar, n_boot, seed, orient, sort, err_style, err_kws, legend, ci, ax, **kwargs)
    635 if not p.has_xy_data:
    636     return ax
--> 638 p._attach(ax)
    640 # Other functions have color as an explicit param,
    641 # and we should probably do that here too
    642 color = kwargs.pop("color", kwargs.pop("c", None))

File [~/.local/lib/python3.10/site-packages/seaborn/_oldcore.py:1255](https://file+.vscode-resource.vscode-cdn.net/home/manon.marchand/Documents/cds-astro-github/tutorials/Notebooks/~/.local/lib/python3.10/site-packages/seaborn/_oldcore.py:1255), in VectorPlotter._attach(self, obj, allowed_types, log_scale)
   1253 # Now actually update the matplotlib objects to do the conversion we want
   1254 grouped = self.plot_data[var].groupby(self.converters[var], sort=False)
-> 1255 for converter, seed_data in grouped:
   1256     if self.var_types[var] == "categorical":
   1257         if self._var_ordered[var]:

File [~/.conda/envs/cds-tutos-dev/lib/python3.10/site-packages/pandas/core/groupby/ops.py:604](https://file+.vscode-resource.vscode-cdn.net/home/manon.marchand/Documents/cds-astro-github/tutorials/Notebooks/~/.conda/envs/cds-tutos-dev/lib/python3.10/site-packages/pandas/core/groupby/ops.py:604), in BaseGrouper.get_iterator(self, data, axis)
    602 splitter = self._get_splitter(data, axis=axis)
    603 keys = self.group_keys_seq
...
    166     out = out.T

File pandas/_libs/algos_take_helper.pxi:1705, in pandas._libs.algos.take_1d_float32_float32()

ValueError: Big-endian buffer not supported on little-endian compiler

While it works perfectly well when reverting to pandas 2.0.0

I tried to have a look in their changelog and there is indeed some changes on dtypes but I'm not sure what the issue might be.

Also, it might be on pandas side?

[mwaskom]

Fairly easy to reproduce in pure pandas

import numpy as np
import pandas as pd
x = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
y = np.array([0.5, 0.6, 0.7], dtype=np.dtype('>f4'))
df = pd.DataFrame({"x": x, "y": y})
list(df.groupby("x"))

Not sure if this should be expected to work or if seaborn is somehow doing something wrong here but based on this probably should be raised upstream.

[ManonMarchand]

Thanks for looking!

This new snippet fails in both versions of pandas while the first one using seaborn only fails for the newest pandas. This is very strange.

On pandas side, this looks like a known issue pandas-dev/pandas#53234

[mwaskom]

AFAICT there's no action for seaborn to take here; whether pandas is able to support this is up to them, but if they're not able to support it, I don't think it's seaborn's business to be changing the endinanness of user input data.