Nonparametric Bayesian Discrete Latent Variable Models for ...
Nonparametric Bayesian Discrete Latent Variable Models for ...
Nonparametric Bayesian Discrete Latent Variable Models for ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
B Mathematical Appendix<br />
B.1 Dirichlet Distribution<br />
In the following, we describe the Dirichlet distribution and give some of its properties<br />
important <strong>for</strong> understanding the Dirichlet processes.<br />
The gamma distribution with shape parameter α ≥ 0 and scale parameter β > 0 is<br />
given as<br />
G(θ | α, β) =<br />
Let θi, i = 1, . . . , k be independent random variables with<br />
1<br />
Γ(α)β α θα−1 exp{−θ/β}. (B.1)<br />
θi ∼ G(αi, 1). (B.2)<br />
The Dirichlet distribution D(α1, . . . , αk) is defined as the distribution of (π1, . . . , πk),<br />
where<br />
k<br />
πi = θi/ θj <strong>for</strong> i = 1, . . . , k. (B.3)<br />
The density function is given as,<br />
j=1<br />
D(π1, . . . , πk | α1, . . . , αk) = Γ( k<br />
j=1 αj)<br />
k<br />
j=1 Γ(αj)<br />
k<br />
j=1<br />
π αj−1<br />
j . (B.4)<br />
Dirichlet distribution is conjugate to the multinomial distribution<br />
A random variable taking two different values with probability π1 and π2 = 1 − π1 is<br />
Bernoulli distributed:<br />
<br />
π1, x = x1<br />
p(x) =<br />
(B.5)<br />
π2, x = x2<br />
The count of the occurrence of one of the events (e.g. x = x1) in a sequence of n<br />
Bernoulli trials is generally represented with the Binomial distribution:<br />
<br />
n<br />
p(y | π1) = Bin(y | π1, n) = π<br />
y<br />
y n−y<br />
1 (1 − π1)<br />
The conjugate prior <strong>for</strong> the Binomial distribution is the Beta distribution:<br />
Γ(α + β)<br />
p(π1) = Beta(π1 | α, β) =<br />
Γ(α)Γ(β) πα−1 1 (1 − π1) β−1 .<br />
119