Families of Parametrized Distributions

Each of the following defines a

  • A sample space \(X\) + sigma algebra \(\mathcal{S}\)
  • A parameter space \(P\)
  • A mapping from the parameter space \(P\) to a subset of the set of measures defined on \((X, \mathcal{S})\)
    • This mapping is typically defined by a probability density together with a reference distribution.
  • The notation \(F(P, Q, ..)\) of application of a function \(F\) to probability distributions \(P\), \(Q\) is defined to be the pushforward by F of the cartesian product of \(P\), \(Q\), ..
\[F_{*}(P \times Q \;\times ...)\]

Bernoulli

Parameters :

  • p \(\in\) \([0,1] \subset \mathbb{R}\)

Sample Space : \(\{0, 1\}\)
Sigma Algebra : Discrete

Measure : \(P(1) = p\) and \(P(0) = 1-p\)

Arithmetic:

  • XOR : \(B(p) \wedge B(q) = B(p (1 - q) + q (1 - p)) = B(q + p - 2qp)\)

  • OR : \(B(p) \vert B(q) = B(p + q - pq)\)

  • AND : \(B(p) \& B(q) = B(pq)\)

  • NOT : \(!B(p) = B(1-p)\)

  • Addition : \(\sum_{i=1}^N B(p) = B(N, p)\)

Links

Beta

Parameters :

  • \(\alpha\) \(\in\) \(\mathbb{R^+}\)

  • \(\beta\) \(\in\) \(\mathbb{R^+}\)

Sample Space : \([0,1] \subset \mathbb{R}\)
Sigma Algebra : Lebesgue

Density : \(\frac{\Gamma(\alpha + \beta)}{\Gamma(\alpha)\Gamma(\beta)}x^{\alpha-1} (1-x)^{\beta -1}\)

Arithmetic:

Links

Binomial

Parameters :

  • N \(\in\) \(\mathbb{N}\)

  • p \(\in\) \([0,1] \subset \mathbb{R}\)

Sample Space : \([0,N] \subset \mathbb{Z}\)
Sigma Algebra : Discrete

Density : \(\rho(n) = {N\choose n} p^n (1-p)^{N-n}\)

Arithmetic:

  • Addition (same p) : \(B(N, p) + B(M, p) = B(N+M, p)\)

Links

Cauchy

Parameters :

Sample Space :
Sigma Algebra :

Arithmetic:

Links

Gamma

Parameters :

  • \(\alpha\) \(\in\) \(\mathbb{R}^+\)

  • \(\beta\) \(\in\) \(\mathbb{R}^+\)

Sample Space : \(\mathbb{R}^+\)
Sigma Algebra : Lebesgue

Density : \(\frac{\beta^\alpha}{\Gamma(\alpha)} x^{\alpha - 1} e^{-\beta x}\)

Arithmetic:

  • Addition : \(Gamma(\alpha, \beta) + Gamma(\alpha', \beta) = Gamma(\alpha + \alpha', \beta)\)

  • Scaling : \(\eta \times Gamma(\alpha, \beta) = G(\alpha, \beta / \eta)\)

Links

Multinomial

Parameters :

  • \(N\) \(\in\) \(\mathbb{N}\)

  • \(p\) \(\in\) \([0,1]^d \subset \mathbb{R}^d\) with \(\sum_i p_i = 1\)

Sample Space :
Sigma Algebra :

Density : \(\rho(n_1,n_2,n_3,..,n_m) = choose(N,n_1,n_2,..,n_m) p_1^{n_1} p_2^{n_2} .. p_m^{n_m}\)

Arithmetic:

Links

Multi-Normal

Parameters :

Sample Space : \(\mathbb{R}^d\)
Sigma Algebra : Legesgue

Density : \(N(\mu, \Sigma) = \frac{1}\{(2\pi)^{d/2} \sqrt{\Sigma} \} \exp[-1/2 (x-\mu)^T\Sigma^{-1}(x-\mu)]\)

Arithmetic:

  • Addition : \(N(\mu, \Sigma) + N(\mu', \Sigma') = N(\mu+\mu', \Sigma+\Sigma')\)

  • Linear Map, with $A$ invertible : \(A \cdot N(\mu, \Sigma) = N(A \mu, A \Sigma A^T)\)

Links

Normal

Parameters :

  • \(\mu\) \(\in\) \(\mathbb{R}\)

  • \(\sigma\) \(\in\) \(\mathbb{R}^+\)

Sample Space : \(\mathbb{R}\)
Sigma Algebra : Lebesgue

Arithmetic:

  • Addition : \(N(\mu, \sigma) + N(\mu', \sigma') = N(\mu+\mu', \sqrt{\sigma^2 + \sigma'^2})\)

  • Scalar Multiplication : \(\lambda N(\mu, \sigma) = N(\lambda \mu, \lambda \sigma)\)

Links

Poisson

Parameters :

  • \(\lambda\) \(\in\) \(\mathbb{R}^+\)

Sample Space : \(\mathbb{N}\)
Sigma Algebra : Discrete

Density : \(\rho(n) = \frac{\lambda^n e^{-n}}{n!}\)

Arithmetic:

  • Sum : \(Pois(\lambda_1)+Pois(\lambda_2) = Pois(\lambda_1 + \lambda_2)\)

Links