python - Why are no colors shown in kde subplots in seaborn pairplot?

Question

I am looking at the iris data set (Fisher 1936). e.g. https://www.kaggle.com/uciml/iris/downloads/Iris.csv

Creating the seaborn pairplot with the arguments

sns.pairplot(iris.drop("Id", axis=1), diag_kind="kde", hue="Species")

return a pairplot with kde charts on the diagonals; However, I am missing the different colors for the different species in the kde plots, the scatters are fine & colorful.

My result is inline with the seaborn docs. http://seaborn.pydata.org/tutorial/axis_grids.html

g = sns.pairplot(iris, hue="species", palette="Set2", diag_kind="kde", size=2.5)

But there a several different examples published showong the colors. e.g. http://www.arunprakash.org/2017/06/data-visualisation-seaborn.html

sns.pairplot(iris, hue='Species', diag_kind='kde', size=2);

or https://www.kaggle.com/benhamner/python-data-visualizations

sns.pairplot(iris.drop("Id", axis=1), hue="Species", size=3, diag_kind="kde")

Has there been a recent change in the seaborn API (ver 0.8.0) ? Have the colors been removed on purpose? Is there a kw to the show them again?

Answer 1

There was an the issue with producing hues on the diagonal of sns.pairplot. This issue is now fixed in version 0.8.1 of seaborn.

In case one is still interested, the following may be a workaround. You may create the underlying PairGrid yourself and map the diagonal and the off_diagonal elements individually. For the diagonal elements, get a color from the current cycler first, then use this color for the kdeplot.

import matplotlib.pyplot as plt
import seaborn as sns
iris = sns.load_dataset("iris")

g =  sns.PairGrid(iris, hue='species', size=2)

def f(x, **kwargs):
    kwargs.pop("color")
    col = next(plt.gca()._get_lines.prop_cycler)['color']
    sns.kdeplot(x, color=col, **kwargs)

g.map_diag(f)
g.map_offdiag(plt.scatter)
g.add_legend()
plt.show()