(ML 17.5) Importance sampling - introduction

[@JadtheProdigy]

But if you cannot draw samples from p, how can you evaluate p(x) in the importance sampling formulation f(x)*(p(x)/q(x))?

[@chandreshmaurya8102]

becoz X is disgtributed according to q and we know that we can sample from q easily. Once we know x, we can compute p(x), the pdf evaluated under p.

[@konstantinkotsev]

That's a cool trick for approximation. Nice explanation too! Thanks!

[@walkingon2001]

thanks, what you said was simple and clear. Looking forward to some examples in the next video.

[@yyalbert4756]

Very clear! Thank you!

[@silvanomarchesi7146]

Really, really good explanation. Thank you!

[@user-pl5zn5jf5d]

What if the p(x) don't have a explicit formula, how do we calculate the value of w(x_i),

[@Paivren]

if you do not know p(x), you cannot come up with a 'good' function q and you cannot calculate the w(x_i) terms.

[@johnnyfrink]

I love your videos! What I didn't get here is: To compute w(x) we do have to compute p, so this must be easy on the one side and p must be known on the other side, incl. partition function, right? And: How do people choose q? Again, we need to know p for that, don't we?

[@evazhao01]

Great video. Thanks.

[@donamincorleone]

Nice video. Helped a lot. Thanks :)

[@maksimilianmusik5995]

awesome video man!

[@achillesarmstrong9639]

wow, very good explanation

[@Daedana]

This was very helpful.

[@shairuno]

Why we want to do the importance sampling when we already know an explicit form of p(x)?

[@tinontentes9854]

Sometimes it's very hard to sample from, so we sample from an easier q(x) instead.

[@ramiBudemaris]

thanks a lot for the video, it was very helpful ... may I ask where is the next video ?

[@cbandle5050]

Great lecture as always. By the way, is there a name for the optimal variance-minimizing choice of q?

[@mathematicsforphysics6882]

thanks

[@danlan4132]

Is it also called injection sampling?

[@junoreactor11]

Do you mean "rejection sampling"?

[@ernestedifor]

I have a simple C# code (below) for evaluating the CDF of an exponentially distributed event using MC. How can I implement IS?
Random rnd = new Random(); //random number generator
double hit = 0, trials = 1000, num=0.9, sim=0;
for (int i = 1; i

[@saravanabalagi]

Find the inverse of the CDF and sample values corresponding to uniform random numbers generated within the bounds. But this is not importance sampling.

[@winoddhamnekar8896]

Sound should be louder.