KERNEL DENSITY ESTIMATION Lecture 3 KERNEL DENSITY ESTIMATION Problems with the Histogram Method The Naïve Estimator Kernel Functions The Kernel Density Estimator Theoretical Properties The Optimal Kernel
Problems with the Histogram Method
Histogram and KDE Simulated Example
The Naïve Estimator
Definition of the Kernel Function
Some Commonly Used Kernel Functions
Examples of Kernels 4 kernels
The Kernel Density Estimator
Examples x=randn(200,1); >> [xpdf,h]=kde0(x);h=.36572 by normal reference method
Examples Undersmooth and oversmooth
Examples Uniform kernel used
Examples Optimal kernel used
Some Remarks
The Convolution Expression
Measures of Performance
Theoretical Properties: Bias and Variance
Theoretical Properties: MSE and the Local Optimal Bandwidth
Theoretical Properties: MISE and the Global Optimal Bandwidth
Some Remarks
The Optimal Kernel
Eff(K) K(t) 1 0.9295 0.9512 0.9859 0.9939 Uniform Gaussian Triangular Kernel Efficiency Efficiencies of some commonly used kernels 1 0.9295 0.9512 0.9859 0.9939 Uniform Gaussian Triangular Biweight Epanechnikov Eff(K) K(t) Kernel 2 1
Example of Kernel Efficiency -1 +1 1
Remarks for Kernel Choice