Setting hue in pairplot on a string column with only 1 item in a category throws exception: ValueError: `dataset` input should have multiple elements.

seaborn 0.9.0, installed via pip.

I have 10 rows, trying to create pairplot. The plot works fine until I set the hue to a string (object) column that has 4 categories with the breakdown of (4, 3, 2, 1). Stack trace below:

ValueError                                Traceback (most recent call last)
<ipython-input-62-cc1e7428015b> in <module>()
      8 # ds.dtypes
      9 # ds
---> 10 sns.pairplot(ds, vars=['mark', 'days', 'hours', 'refs'], hue='topic')
     11 # dataset.iloc[0:10,[8,19]]
     12 

/Users/piglet/Library/Python/2.7/lib/python/site-packages/seaborn/axisgrid.pyc in pairplot(data, hue, hue_order, palette, vars, x_vars, y_vars, kind, diag_kind, markers, height, aspect, dropna, plot_kws, diag_kws, grid_kws, size)
   2109             diag_kws.setdefault("shade", True)
   2110             diag_kws["legend"] = False
-> 2111             grid.map_diag(kdeplot, **diag_kws)
   2112     
   2113     # Maybe plot on the off-diagonals

/Users/piglet/Library/Python/2.7/lib/python/site-packages/seaborn/axisgrid.pyc in map_diag(self, func, **kwargs)
   1397                     color = fixed_color
   1398                 
-> 1399                 func(data_k, label=label_k, color=color, **kwargs)
   1400             
   1401             self._clean_axis(ax)

/Users/piglet/Library/Python/2.7/lib/python/site-packages/seaborn/distributions.pyc in kdeplot(data, data2, shade, vertical, kernel, bw, gridsize, cut, clip, legend, cumulative, shade_lowest, cbar, cbar_ax, cbar_kws, ax, **kwargs)
    689         ax = _univariate_kdeplot(data, shade, vertical, kernel, bw,
    690                                  gridsize, cut, clip, legend, ax,
--> 691                                  cumulative=cumulative, **kwargs)
    692     
    693     return ax

/Users/piglet/Library/Python/2.7/lib/python/site-packages/seaborn/distributions.pyc in _univariate_kdeplot(data, shade, vertical, kernel, bw, gridsize, cut, clip, legend, ax, cumulative, **kwargs)
    292                               "only implemented in statsmodels."
    293                               "Please install statsmodels.")
--> 294         x, y = _scipy_univariate_kde(data, bw, gridsize, cut, clip)
    295 
    296     # Make sure the density is nonnegative

/Users/piglet/Library/Python/2.7/lib/python/site-packages/seaborn/distributions.pyc in _scipy_univariate_kde(data, bw, gridsize, cut, clip)
    364     """Compute a univariate kernel density estimate using scipy."""
    365     try:
--> 366         kde = stats.gaussian_kde(data, bw_method=bw)
    367     except TypeError:
    368         kde = stats.gaussian_kde(data)

/Users/piglet/Library/Python/2.7/lib/python/site-packages/scipy/stats/kde.pyc in __init__(self, dataset, bw_method)
    167         self.dataset = atleast_2d(dataset)
    168         if not self.dataset.size > 1:
--> 169             raise ValueError("`dataset` input should have multiple elements.")
    170 
    171         self.d, self.n = self.dataset.shape

ValueError: `dataset` input should have multiple elements.

[adam-ah]

I wouldn’t agree with this statement.

I must have misread your argument because surely you aren’t saying that a plotting library should throw a random exception from a third party library because it was asked to color in dots?

You can throw a warning, saying that the diagram diagonal isn’t using colours, because of whatever reason.

You can throw an error, saying that ‘kde’ diagonal requires at least two elements in each group (for whatever reason), and say that you can use the ‘hist’ parameter to draw a histogram instead of kde.

But what you cannot do is to let a third party library throw an arbitrary exception because seaborn didn’t check the required parameters the library needs to do its thing.

The issue here isn’t with seaborn, the issue is with the underlying math.

Just like we cannot blame calling toString() on a null value, or blame sqrt() when passing in a negative value, we cannot blame a third party library when we are feeding in incorrect parameters.

EDIT: With the .pairplot API the user never actually asked for a kde plot, only for a pair-plot, and seaborn decided to use a kde plot; it’s difficult to reason that it is a user error either. The data can be plotted and colored in, so overall it would be hard to conclude that it is not a bug.

[mwaskom]

Well, you’ve failed in that aim, but succeeded in irritating me to the point that I’ve moved this issue to the bottom of the stack of things I choose to spend time working on. Congrats!