R Tutorial 28: Acceptance Rejection Method for Sampling (Envelope a Distribution)

[@wilsonzhang5258]

Very clear and intuitive video! Thanks for helping me with my computational statistics exam review!

[@mattulrich2522]

^^^ Agreed

[@chandrap29]

yeah it is very useful for basics of accept reject method. thanks a lot. keep on other videos.

[@AllenKei]

Thank you very much!

[@SebasTricht]

Great video, really clear explanations. Thanks!

[@giacomobonomelli]

Thank you!! very helpful video and great explanation by the way!

[@Johnnyboycurtis]

A t-distribution or Laplace distribution would be good candidates for this. However, for tail probabilities you may need to use importance sampling. Good video!

[@AllenKei]

Thank you so much for your great comments!!! Allen

[@victorkalatzis8101]

Hello! How can take sample from N(0,1) distribution truncated in [-1,2] with envelope Distribution laplace(0,b) ? And how can find an envelope distribution for N(0,1) that it has the biggest probability of acceptance? Thanks

[@user-xt9js1jt6m]

very helpful sir!!

[@tatianacalyamador7389]

Nice video. I have a question... How can I use this method if I have a target density N(2,4)?

[@mmski8910]

How do I choose the envelope distribution. Should I go by trial and error? What if I graph the target dist and say it looks like an exp decay, if i use an exp envelope and it doesn’t work, where do I go from there?

[@nicolaspitsillides4172]

why do you have rep(0,100000)? Could you have another number instead of 0?

[@orcunsami1450]

You found Integrate as a 2 but you have never used it. What am I missing? For instance, in the loop, f has not been changed to pdf as how you are told.

[@orcunsami1450]

I liked the video btw, thanks

[@orcunsami1450]

I divided f function to 2 and result did not change. Maybe he did for this reason, not sure.

[@rafiqulislam1085]

Excellent

[@jdlopez131]

pdf of normal , the constant part is 1/sqrt(2*pi) . It is not sqrt(2/pi)

[@AllenKei]

f(x) is not the normal distribution. You can try to take the integral of f(x), and find its normalizing constant. Let me know if that is equal to 1/sqrt(2*pi).

[@jdlopez131]

Don't you stare in second 30 that you are speaking about the pdf of the normal distribution?

[@jdlopez131]

If you integrate the real normal distribution it will get you 1...as it is supposed to be

[@jdlopez131]

You integrate that function, like you did in your video, and you got 2 instead (what was that integral supposed to prove?)

[@ltc_dingzhen_zood]

多谢老哥

[@jacobm7026]

So this works to find the posterior then?

[@AllenKei]

Hi, Thanks for your question. This is just to generate some samples data that follows a specific distribution. It is not yet related to Bayesian yet. Allen

[@jacobm7026]

@@AllenKei Ok thank you! I always just thought rejection sampling is a way to find the posterior distribution through sampling . Looking forward to the Bayesian stuff!