Problem statement:
Given observational data, estimate the underlying probability
distribution.
Introduction
Check also this
very gentle introduction to KDE.
Thus, we have a list of data points, by what distribution this data
would be generated?
Kernel estimation puts a kernel (for example a Gaussian distribution)
on every point and sums it to construct the overall distribution.
Example applet
The rationale is that since only a limited number of points is
available, every point is partly uncertain.