How to efficiently generate a random integer within a range from an arbitrary probability distribution
조회 수: 2 (최근 30일)
이전 댓글 표시
I need to generate a random integer within a range from an arbitrary probability distribution, within a loop of 100000 iterations. My implementation works, but I am not sure it is mathematically clean, and it takes forever:
pdf = [ 0.9 0.3 0.003 0.1 0.07 0.0005 0.003 0.15 0.009 0.08 ]; % discrete prob distrib function
cdf = cumsum(pdf); % cumulative distribution function
cdf = cdf / max(cdf); cdf(1) = 0; % normalization
index = ceil(interp1(cdf, [1:numel(pdf)], rand(1)))
Notice that the pdf above is just an example: my actual case is a vector of about 500 numbers.
Here is a different solution, which seems mathematically cleaner, but does not work for my overall problem, and is just as slow as above:
pdf = [ 0.9 0.3 0.003 0.1 0.07 0.0005 0.003 0.15 0.009 0.08 ]; % discrete prob distrib function
cdf = cumsum(pdf); % cumulative distribution function
cdf = cdf - min(cdf); cdf = cdf / max(cdf); % normalization
index = round(interp1(cdf, [1:numel(pdf)], rand(1)))
Is there a more efficient/correct way to do this?
댓글 수: 0
채택된 답변
the cyclist
2017년 1월 8일
편집: the cyclist
2017년 1월 8일
I think that
index = sum(rand()>cdf)+1;
will be much faster than using interp1 as you do, and will give the same result.
댓글 수: 4
the cyclist
2017년 1월 8일
Regarding speed ...
I get roughly equivalent results between my solution and randsample. Note that both of these solutions can be vectorized:
index = sum(rand(N,1)>cdf,2)';
and
index = randsample(1:numel(pdf),N,true,pdf);
These will both generate N = 100,000 samples in a few milliseconds.
I did not try to vectorize your solution, but as it currently stands, it took 5 seconds to generate that many.
추가 답변 (1개)
the cyclist
2017년 1월 8일
Do you have the Statistics and Machine Learning Toolbox? If so, you can do this with the randsample command.
참고 항목
카테고리
Help Center 및 File Exchange에서 Creating and Concatenating Matrices에 대해 자세히 알아보기
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!