0
$\begingroup$

Given is following function which is also a density

$$f(x)=\left\{\begin{matrix} \frac{3}{4}(2x-x^2) \text{ if } x \in (0,2)\\ 0 \,\,\,\,\,\,\,\text{ else } \end{matrix}\right.$$

Create a random variable with this density.

I ask this question because they can ask similar in exam and I like to know how to do it good and correct?

I think what would be good to use here is rejection sampling. I read about it on Wikipedia but not sure how use it here but nothing else comes to my mind to solve it.

Formula I find is $$M := \sup_{x}\frac{f(x)}{g(x)} < \infty$$

I sum up algorithm because on Wikipedia it's very long:

  1. Simulate $U \sim R(0,1)$, in case you don't know this notation: It means that random variable $U$ is equally / uniformly distributed from interval $0$ to $1$
  2. Simulate $Y \sim g$
  3. Accept $X=Y$ if $U \leq \frac{1}{M} \frac{f(Y)}{g(Y)}$, else go to step 1. (repeat it)

  1. We have $P(X<x) = \int_{0}^{x} dt = x$ so $X$ has density function

$$f(x)=\left\{\begin{matrix} 1 \text{ if }0 \leq x < 1\\ 0 \text { else } \end{matrix}\right.$$

I hope this is correct for first step? But I don't know how to do second step and that's why third I can't do too : /

Maybe there is other method for solving this? Because I only know about this.

$\endgroup$
2
$\begingroup$

I'd advise you to read some practical examples on how to do rejection sampling.

Just wanted to say, the CDF ($CDF(x)=\int_0^xf(x')dx'$) of your probability function can actually be inverted. And if you fill in your $x\sim~Uniform[0,1]$ into the inverted CDF, you get a random variable distributed like your $f(x)$. The method is explained here: https://en.wikipedia.org/wiki/Inverse_transform_sampling

$\endgroup$
  • $\begingroup$ Yes, I agree ; in such a case, it's inverse transform sampling $F^{-1}(U)$ where $U$ is uniform on $[0,1] $ and $F$ a primitive function = cdf $F$ of pdf $f$ that will be the most simple/efficient. $\endgroup$ – Jean Marie Jan 23 '18 at 22:34

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.