ValueError: The palette list has the wrong number of colors.

Given a DataFrame with two categories:

df = pd.DataFrame({"a": [1,2,3,4,5,6,7,8,9,10], "b": ["a", "b", "a", "b", "a", "b", "a", "b", "a", "b", ]})

When I provide a palette with lineplot, I get an error:

sns.lineplot(data=df, x="a", y="a", hue="b", palette=sns.color_palette())

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-13-4e6a1a2e63cf> in <module>
----> 1 sns.lineplot(data=df, x="a", y="a", hue="b", palette=sns.color_palette())

/data1/mschroeder/miniconda3/envs/20-assdc/lib/python3.7/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 

/data1/mschroeder/miniconda3/envs/20-assdc/lib/python3.7/site-packages/seaborn/relational.py in lineplot(x, y, hue, size, style, data, palette, hue_order, hue_norm, sizes, size_order, size_norm, dashes, markers, style_order, units, estimator, ci, n_boot, seed, sort, err_style, err_kws, legend, ax, **kwargs)
    696     )
    697 
--> 698     p.map_hue(palette=palette, order=hue_order, norm=hue_norm)
    699     p.map_size(sizes=sizes, order=size_order, norm=size_norm)
    700     p.map_style(markers=markers, dashes=dashes, order=style_order)

/data1/mschroeder/miniconda3/envs/20-assdc/lib/python3.7/site-packages/seaborn/_core.py in map(cls, plotter, *args, **kwargs)
     51         # This method is assigned the __init__ docstring
     52         method_name = "_{}_map".format(cls.__name__[:-7].lower())
---> 53         setattr(plotter, method_name, cls(plotter, *args, **kwargs))
     54         return plotter
     55 

/data1/mschroeder/miniconda3/envs/20-assdc/lib/python3.7/site-packages/seaborn/_core.py in __init__(self, plotter, palette, order, norm)
    117                 cmap = norm = None
    118                 levels, lookup_table = self.categorical_mapping(
--> 119                     data, palette, order,
    120                 )
    121 

/data1/mschroeder/miniconda3/envs/20-assdc/lib/python3.7/site-packages/seaborn/_core.py in categorical_mapping(self, data, palette, order)
    202                 if len(palette) != n_colors:
    203                     err = "The palette list has the wrong number of colors."
--> 204                     raise ValueError(err)
    205                 colors = palette
    206             else:

ValueError: The palette list has the wrong number of colors.

With pointplot, this works fine (it uses the first two colors of the provided palette):

sns.pointplot(data=df, x="a", y="a", hue="b", palette=sns.color_palette())

I think the behavior of pointplot is the correct one.

[mwaskom]

I definitely agree that it would be ideal if all functions used the same rules for handling (and error-checking) their inputs. Currently lineplot and pointplot don’t go through the same code path, and the path forward there is to refactor the internals of the categorical plots so that they can (#2429).

As for how to handle cases like these, I think I am currently leaning towards saying that there should be an informative warning when a len(palette) != len(hue_levels), rather than an error. That will need to be the case anyway for the categorical plots, at least temporarily, because we can’t just flip a switch and make inputs that used to be happily accepted error (unless a strong case could be made that that behavior was buggy).

[mwaskom]

But why argue?

Because there is always more to do in open source than there is time to do it, so it is important to be clear for the purposes of prioritization. Also, because issues that confuse “this is incorrect because it does not work as intended” and “this is incorrect because it does not work as i would prefer” are an irritation that, over time, drains one’s interest in working on any of them.

In any case,

And here, the the palette behavior of pointplot is a superset of the behavior of lineplot, but much more flexible.

I don’t agree. First, I think you’re overstating the apparent limitation. You’re already making a custom list, all you need to do is add one parameter so that it has the correct number of colors.

Second, the error is strict but also prevents you from making unintended mistakes. In your particular case it appears that you are not at risk of problems but if you were passing a color list that had fewer distinct values than you needed, it would have to cycle and you’d get a plot that can’t be clearly deciphered. And if you were using a gradient color palette, taking the first few colors would give you less contrast than you’d get if you generated the palette with the right length to begin with:

f, axs = plt.subplots(1, 2)
pal = "viridis"
sns.lineplot(data=tips, x="day", y="total_bill", hue="smoker", palette=sns.color_palette(pal)[:2], ax=axs[0])
sns.lineplot(data=tips, x="day", y="total_bill", hue="smoker", palette=pal, ax=axs[1])
f.tight_layout()

Again I am open to the idea that the error is too constraining and that this should be at most a warning. But there are considerations that you might not have thought of because they don’t bear on your very specific usecase.