relplot gives a cryptic error message with `col='size'`

[jhncls]

In seaborn 0.11.1:

import seaborn as sns
tips = sns.load_dataset('tips')
sns.relplot(x='total_bill', y='tip', col='size', data=tips)

Generates "ValueError: could not broadcast input array from shape (244,2) into shape (244,)"
col='size' doesn't seem to be interpreted as a column name.

[mwaskom]

Thanks for reporting. Here's the full traceback that I see:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/_core.py in categorical_order(vector, order)
   1471             try:
-> 1472                 order = vector.cat.categories
   1473             except (TypeError, AttributeError):

~/miniconda3/envs/py39/lib/python3.9/site-packages/pandas/core/generic.py in __getattr__(self, name)
   5461                 return self[name]
-> 5462             return object.__getattribute__(self, name)
   5463 

AttributeError: 'DataFrame' object has no attribute 'cat'

During handling of the above exception, another exception occurred:

AttributeError                            Traceback (most recent call last)
~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/_core.py in categorical_order(vector, order)
   1475                 try:
-> 1476                     order = vector.unique()
   1477                 except AttributeError:

~/miniconda3/envs/py39/lib/python3.9/site-packages/pandas/core/generic.py in __getattr__(self, name)
   5461                 return self[name]
-> 5462             return object.__getattribute__(self, name)
   5463 

AttributeError: 'DataFrame' object has no attribute 'unique'

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-15-c57b6168b51b> in <module>
      5 
      6 #Plotting our subplots dividing with smoker
----> 7 sns.relplot(x = "total_bill", y = "tip", hue = "smoker", col = "size", data = tips)
      8 
      9 #Showing our plots in a proper format

~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/_decorators.py in inner_f(*args, **kwargs)
     44             )
     45         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 46         return f(**kwargs)
     47     return inner_f
     48 

~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/relational.py in relplot(x, y, hue, size, style, data, row, col, col_wrap, row_order, col_order, palette, hue_order, hue_norm, sizes, size_order, size_norm, markers, dashes, style_order, legend, kind, height, aspect, facet_kws, units, **kwargs)
   1013     facet_kws = {} if facet_kws is None else facet_kws.copy()
   1014     facet_kws.update(grid_kws)
-> 1015     g = FacetGrid(
   1016         data=full_data,
   1017         col_wrap=col_wrap, row_order=row_order, col_order=col_order,

~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/_decorators.py in inner_f(*args, **kwargs)
     44             )
     45         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 46         return f(**kwargs)
     47     return inner_f
     48 

~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/axisgrid.py in __init__(self, data, row, col, hue, col_wrap, sharex, sharey, height, aspect, palette, row_order, col_order, hue_order, hue_kws, dropna, legend_out, despine, margin_titles, xlim, ylim, subplot_kws, gridspec_kws, size)
    334             col_names = []
    335         else:
--> 336             col_names = categorical_order(data[col], col_order)
    337 
    338         # Additional dict of kwarg -> list of values for mapping the hue var

~/miniconda3/envs/py39/lib/python3.9/site-packages/seaborn/_core.py in categorical_order(vector, order)
   1476                     order = vector.unique()
   1477                 except AttributeError:
-> 1478                     order = pd.unique(vector)
   1479 
   1480                 if variable_type(vector) == "numeric":

~/miniconda3/envs/py39/lib/python3.9/site-packages/pandas/core/algorithms.py in unique(values)
    395     array([('a', 'b'), ('b', 'a'), ('a', 'c')], dtype=object)
    396     """
--> 397     values = _ensure_arraylike(values)
    398 
    399     if is_extension_array_dtype(values):

~/miniconda3/envs/py39/lib/python3.9/site-packages/pandas/core/algorithms.py in _ensure_arraylike(values)
    239             if isinstance(values, tuple):
    240                 values = list(values)
--> 241             values = construct_1d_object_array_from_listlike(values)
    242         else:
    243             values = np.asarray(values)

~/miniconda3/envs/py39/lib/python3.9/site-packages/pandas/core/dtypes/cast.py in construct_1d_object_array_from_listlike(values)
   1636     # making a 1D array that contains list-likes is a bit tricky:
   1637     result = np.empty(len(values), dtype="object")
-> 1638     result[:] = values
   1639     return result
   1640 

ValueError: could not broadcast input array from shape (244,2) into shape (244)

[mwaskom]

OK I think the answer is that it's a bug caused by the external column name colliding with one of the internal variable names (because you can do relplot(size=...). I haven't tracked it down completely but I suspect this has to do with the relatively complex relplot-specific logic of joining the internal data structure with the faceting columns.

I would like to simplify that logic in general so I am hoping this bug will get fixed indirectly as part of that process.

Current workaround is to rename the column something that doesn't collide with the seaborn variable names:

sns.relplot(x="total_bill", y="tip", hue="smoker", col=tips["size"].rename(" size"), data=tips)