This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
world:ratio-of-uniform [2023/04/05 17:41] rdouc |
world:ratio-of-uniform [2023/04/20 08:57] (current) rdouc [The idea] |
||
---|---|---|---|
Line 1: | Line 1: | ||
{{page>:defs}} | {{page>:defs}} | ||
+ | ====== The ratio of uniform method ====== | ||
- | If $ (X,U) \sim \unif\lrb{(x,y); 0\leq x\leq f(y)}$ | + | The rejection algorithm is based on the following property: |
+ | * $ (X,Y) \sim \unif\set{(x,y)}{0\leq x\leq M f(y)}$ if and only if $Y \sim f$ and $X|_{Y=y} \sim \unif[0,Mf(y)]$. | ||
- | * [[https://dl.acm.org/doi/pdf/10.1145/355744.355750|ratio of uniform]] | + | The idea of the ratio-of-uniform method is based on the following property: if $ (U,V) \sim \unif\set{(u,v)}{0\leq u\leq \sqrt{M f(v/u)}}$, then $V/U \sim f$. This can be shown from the change of variable $x=u$, $y=v/u$, i.e. $u=x$, $v=xy$. |
- | * {{ :mynotes:bf01889987.pdf | Generalisation}} | + | |
+ | A simple generalisation of this result is: if $ (U,V) \sim \unif\set{(u,v)}{0\leq u\leq G^{-1}\lr{M f\lr{\frac { v} {g(u)}}}}$, then $V/g(U) \sim f$ where $g: \rset^+ \to \rset^+_*$ and $G(x)=\int_0^x g(u) \rmd u$. | ||
+ | |||
+ | As far as I can see, these methods can only be interesting if $V=Y g(X)$ or $U=X$ are easy to simulate when $Y \sim f$ and $X|_Y \sim \unif\lrcb{[0,f(Y)]}$. This is very linked to rejection algorithm... | ||
+ | |||
+ | * [[https://dl.acm.org/doi/pdf/10.1145/355744.355750|ratio-of-uniform method]] | ||
+ | * {{ :mynotes:bf01889987.pdf | Generalisation of the ratio-of-uniform method}} |