Seaborn: which type of kernel density extimation

Question

Setting seaborn.joinplot's parameter kind = 'kde' will plot the joint distribution kernel density estimations as e.g. shown in the examples sections:

From my observations, it does a very good job, but seaborn's documentation does not specify the type of kde (e.g. Gaussian?? from my visual observations it does not seem to be gaussian... but maybe it is?)

What is the kernel density estimation method seaborn's jointplot uses? Is there any way to get a handler to it and estimate the density at specific points?

Answer 1

Depending on whether you have statsmodels installed, seaborn will use

• statsmodels.nonparametric.api.KDEMultivariate
  (see here for the seaborn source)

else

• scipy.stats.gaussian_kde
  (see here for the seaborn source)

There is no way to get a handle to those from within seaborn. You may however use those function directly in your code and later plot a matplotlib contourf plot of the result.