2
$\begingroup$

Slice Sampling asks to draw uniformly from $f^{-1}]y,+\infty[$. Wikipedia page

However, how can we be sure that a uniform defined over the set $f^{-1}]y,+\infty[$ is in fact proper?

If I had to guess, I would say this set must be compact... is this intuition correct?

$\endgroup$
  • 1
    $\begingroup$ +1. We might suppose $f$ is assumed to be at least piecewise continuous. But even if $f$ is infinitely differentiable, the sets $f^{-1}\left(y,\infty\right)$ need not be compact. Compactness isn't essential--but having finite Lebesgue measure is. The Wikipedia language about unimodality provides some clues concerning the additional implicit assumptions being made. $\endgroup$ – whuber Aug 2 '18 at 14:02
2
$\begingroup$

The slice sampler for a target proportional to $f$ goes by

  1. simulating $u^{t+1}\sim\mathbb{I}_{(0,f(x^t))}(u)$
  2. simulating $x^{t+1}\sim\mathbb{I}_{\{x;\,f(x)>u^{t+1}\}}(x)$

and both sets have finite measures, since $f(x^t)<\infty$ and $$\mu(\{x;\,f(x)>u^{t+1}\})=\int_{\{x;\,f(x)>u^{t+1}\}}\text{d}\mu(x)<\int_{ \{x;\,f(x)>u^{t+1}\} }\dfrac{f(x)}{u^{t+1}}\text{d}\mu(x)\le \frac{1}{u^{t+1}}$$ hence the Uniform distributions are well-defined.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.