Divergência de Kullback-Leibler entre duas distribuições gama

Optando por parametrizar a distribuição gama $\Gamma(b,c)$ pelo pdf $g(x;b,c) = \frac{1}{\Gamma(c)}\frac{x^{c-1}}{b^c}e^{-x/b}$ A divergência de Kullback-Leibler entre$\Gamma(b_q,c_q)$e$\Gamma(b_p,c_p)$é dada por [1] como

$$\begin{align} KL_{Ga}(b_q,c_q;b_p,c_p) &= (c_q-1)\Psi(c_q) - \log b_q - c_q - \log\Gamma(c_q) + \log\Gamma(c_p)\\ &\qquad+ c_p\log b_p - (c_p-1)(\Psi(c_q) + \log b_q) + \frac{b_qc_q}{b_p} \end{align}$$

Estou supondo que $\Psi(x):= \Gamma'(x)/\Gamma(x)$ é a função digamma .

Isso é dado sem derivação. Não consigo encontrar nenhuma referência que derive isso. Qualquer ajuda? Uma boa referência seria suficiente. A parte difícil é integrar o em um pdf gama.$\log x$

[1] WD Penny, KL-Divergences of Densities Normal, Gamma, Dirichlet e Wishart , disponível em: www.fil.ion.ucl.ac.uk/~wpenny/publications/densities.ps

A divergência KL é uma diferença de integrais da forma

$$ \ eqalign {I (a, b, c, d) & = \ int_0 ^ {\ infty} \ log \ left (\ frac {e ^ {- x / a} x ^ {b-1}} {a ^ b \ Gamma (b)} \ right) \ frac {e ^ {- x / c} x ^ {d-1}} {c ^ d \ Gamma (d)} dx \

& = - \ frac {1} {a} \ int_0 ^ \ infty \ frac {x ^ de ^ {- x / c}} {c ^ d \ Gamma (d)} \, dx - \ log (a ^ b \ Gamma (b)) \ ​​int_0 ^ \ infty \ frac {e ^ {- x / c} x ^ {d-1}} {c ^ d \ Gamma (d)} \, dx \ & \ quad + (b- 1) \ int_0 ^ \ infty \ log (x) \ frac {e ^ {- x / c} x ^ {d-1}} {c ^ d \ Gamma (d)} \, dx \

&=-\frac{cd}{a} - \log(a^b\Gamma(b)) + (b-1)\int_0^\infty \log(x) \frac{e^{-x/c}x^{d-1}}{c^d\Gamma(d)}\,dx }$$

We just have to deal with the right hand integral, which is obtained by observing

$$\eqalign{ \frac{\partial}{\partial d}\Gamma(d) =& \frac{\partial}{\partial d}\int_0^{\infty}e^{-x/c}\frac{x^{d-1}}{c^d}dx\\ =& \frac{\partial}{\partial d} \int_0^\infty e^{-x/c} \frac{(x/c)^{d-1}}{c}\,dx\\ =&\int_0^\infty e^{-x/c}\frac{x^{d-1}}{c^d} \log\frac{x}{c} \,dx\\ =&\int_0^{\infty}\log(x)e^{-x/c}\frac{x^{d-1}}{c^d}dx - \log(c)\Gamma(d). }$$

Whence

$$\frac{b-1}{\Gamma(d)}\int_0^{\infty} \log(x)e^{-x/c}(x/c)^{d-1}dx = (b-1)\frac{\Gamma'(d)}{\Gamma(d)} + (b-1)\log(c).$$

Plugging into the preceding yields

$$I(a,b,c,d)=\frac{-cd}{a} -\log(a^b\Gamma(b))+(b-1)\frac{\Gamma'(d)}{\Gamma(d)} + (b-1)\log(c).$$

The KL divergence between $\Gamma(c,d)$ and $\Gamma(a,b)$ equals $I(c,d,c,d) - I(a,b,c,d)$, which is straightforward to assemble.

---

Implementation Details

Gamma functions grow rapidly, so to avoid overflow don't compute Gamma and take its logarithm: instead use the log-Gamma function that will be found in any statistical computing platform (including Excel, for that matter).

The ratio $\Gamma^\prime(d)/\Gamma(d)$ is the logarithmic derivative of $\Gamma,$ generally called $\psi,$ the digamma function. If it's not available to you, there are relatively simple ways to approximate it, as described in the Wikipedia article.

Here, to illustrate, is a direct R implementation of the formula in terms of $I$. This does not exploit an opportunity to simplify the result algebraically, which would make it a little more efficient (by eliminating a redundant calculation of $\psi$).

#
# `b` and `d` are Gamma shape parameters and
# `a` and `c` are scale parameters.
# (All, therefore, must be positive.)
#
KL.gamma <- function(a,b,c,d) {
  i <- function(a,b,c,d)
    - c * d / a - b * log(a) - lgamma(b) + (b-1)*(psigamma(d) + log(c))
  i(c,d,c,d) - i(a,b,c,d)
}
print(KL.gamma(1/114186.3, 202, 1/119237.3, 195), digits=12)

The Gamma distribution is in the exponential family because its density can be expressed as:

$$\begin{align} \newcommand{\mbx}{\mathbf{x}} \newcommand{\btheta}{\boldsymbol{\theta}} f(\mbx \mid \btheta) &= \exp\bigl(\eta(\btheta) \cdot T(\mbx) - g(\btheta) + h(\mbx)\bigr) \end{align}$$

Looking at the Gamma density function, its log-normalizer is

$$g(\btheta) = \log(\Gamma(c)) + c\log(b)$$ with natural parameters $$\btheta = \left[\begin{matrix}c-1\\-\frac1 b\end{matrix}\right]$$

All distributions in the exponential family have KL divergence:

$$\begin{align} KL(q; p) &= g(\btheta_p) - g(\btheta_q) - (\btheta_p-\btheta_q) \cdot \nabla g(\btheta_q). \end{align}$$

There's a really nice proof of that in:

Frank Nielsen, École Polytechnique, and Richard Nock, Entropies and cross-entropies of exponential families.