Performance of random number generator
5 views (last 30 days)
Show older comments
I'm trying to generate a lognormal random number that is truncated, and the only way that I've seen that I can do that is by first fitting the distribution to the data: pd = fitdist(data,'lognormal');
then truncating it: t = truncate(pd,minw,maxw);
and then creating the random numbers with: r = random(t,1,10000);
The problem I have is that I'm doing this for 100 different values of mu-sigma-pairs, and doing 10,000 simulations, and it is taking forever. I started running the code this morning (10 am) in Matlab R2014b in a server and 8 hours later it is still not done.
Before, I was doing this without truncating the values by using R = lognrnd(mu,sigma), and it only took 1.5 to 2 hours to run the exact same code.
What can I do to make it faster? Is there a way to truncate the random number while using lognrnd?
Thanks for your help
7 Comments
Answers (1)
Walter Roberson
on 5 Oct 2015
total_samples_needed = 10000;
have_samples = [];
while true
num_needed_now = total_samples_needed - length(have_samples);
if num_needed_now <= 0; break; end
current_samples = RANDOMGENERATOR(1,num_needed_now);
current_samples(current_samples < Lower_Bound | current_samples > Upper_bound) = [];
have_samples = horzcat(have_samples, current_samples);
end
If you want to make it more efficient, you can ask it to generate 1.1 (or as appropriate) times num_needed_now and at the end discard any unneeded ones you generated. An appropriate multiplication factor would be 1 divided by the cdf between Lower_Bound and Upper_Bound.
RANDOMGENERATOR would be replaced by the appropriate call for your purposes.
2 Comments
Walter Roberson
on 5 Oct 2015
Just have the current mu and sigma passed to RANDOMGENERATOR . As long as they do not have to change within a run of total_samples_needed there is no problem.
See Also
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!