MATLAB Answers

MByk
0

Different Bayes Optimization Results

Asked by MByk
on 7 Feb 2019
Latest activity Edited by MByk
on 2 Apr 2019
Sorry if it is a dumb question but is it normal to get different values despite using the same seed value? Or Am I making a mistake? I am using rng(1234,'twister') as seed value and Bayesian optimization in parallel. Thanks for the help.

  1 Comment

Hard to tell without seeing sufficient code. Probable user error. :)
Using rng as you did will reset the seed. Thus we see:
rng(1234,'twister')
>> rand
ans =
0.19152
>> rand
ans =
0.62211
>> rng(1234,'twister')
>> rand
ans =
0.19152
So when I use rng again, the sequence repeats. Perhaps you are doing something along the line to interrupt that sequence, thus perhaps a rng('shuffle") is done somewhere, perhaps by some other code you are using, even something you may have downloaded. As I said, hard to know.

Sign in to comment.

1 Answer

Answer by Alan Weiss
on 7 Feb 2019
 Accepted Answer

Sorry, it is not clearly documented, but when running Bayesian optimization in parallel, results are not reproducible. You can see this from the Parallel Bayesian Algorithm description, where it discusses what happens if too many workers are idle. The algorithm can perform differently depending on the state of the workers, so is not reproducible.
Alan Weiss
MATLAB mathematical toolbox documentation

  2 Comments

Thanks for the answer sir. So nothing wrong with the code but for reproducibility I must shut down the parallel pool.
So even in MATLAB, idle workers are a problem. ;-)

Sign in to comment.