(a) (b)

A comparison between the kernel approach and the semi-parametric approach

ng a density function for a data set. The thick lines stand for the final densities.

nes stand for the kernel/component densities. (a) The Kernel-based model. (b)

arametric model.

using the semi-parametric approach to estimate a density

for a data set, it is important to make a right assumption of the

f basic densities or components. In other words, the component

hould be accurately decided. Figure 2.14(a) shows a case, where

component number was three, but the designed component

was two. The resulting density is certainly misleading. However,

ploying too many basic densities, this problem may disappear

ome components may be duplicated at all. Figure 2.14(b) shows

here a semi-parametric model employing too many components

set which was actually a mixture of two Gaussians still worked.

el was therefore heavier.

(a) (b)

a) The estimated density with too few components. (b) The estimated density

any components.