2h2 h just likes the standard deviation of the Gaussian components. The density model is obtained by placing a Gaussian over each data point and then adding up the contributions over the whole data set,and then dividing by N,so that the density is correctly normalized. We can choose any other kernel function k(a)subject to the following conditions(即满足成为概率分布的条件)： ●k(u≥O,nonnegative k(u)=k(-u),symmetry k(u)du 1,normalized 16/27p(x) = 1 N ∑ N n=1 1 (2πh 2 ) D/2 exp { − (x − xn) T (x − xn) 2h 2 } h just likes the standard deviation of the Gaussian components. ▶ The density model is obtained by placing a Gaussian over each data point and then adding up the contributions over the whole data set, and then dividing by N , so that the density is correctly normalized. ▶ We can choose any other kernel function k(u) subject to the following conditions (即满足成为概率分布的条件): 1 k(u) ≥ 0, nonnegative 2 k(u) = k(−u), symmetry 3 ∫ k(u)du = 1, normalized 16 / 27