how to throttle CPU usage for fitrgp?

2 次查看（过去 30 天）
jessupj
jessupj 2023-12-14
评论： Drew 2023-12-15
(I hate asking something this trivial, but I'm frustrated... thanks for reading.)
My gaussian process regression via fitrgp automatically uses the maximum number of physical cores.
maxNumCompThreads(1)
mdl = fitrgp(X_train, Y_train,'KernelFunction', 'squaredexponential');
Specifically, setting the maxNumCompThreads is not effective in limiting CPU usage.
What am i missing here for throttling built-ins? I want to hard-limit it to may maybe 20 or 30 cores on this shared memory system.

请先登录，再回答此问题。

采纳的回答

jessupj
jessupj 2023-12-14
移动：Walter Roberson 2023-12-15
(SOLVED)
So... for whatever reason, restarting matlab got maxNumCompThreads to start managing usage. (Same system environment, same workspace definitions). I think the explanation/answer here is 'something broke internally; restart the matlab kernel to fix it' unless there's a better answer that doesn't require that.
  1 个评论
Drew
Drew 2023-12-15
On what platform & CPU configuration did you observe reaching a state where maxNumCompThreads is not managing max thread usage? Please file a support case with reproduction steps if the condition is repeatable.

请先登录，再进行评论。

更多回答（1 个）

Drew
Drew 2023-12-14
Use maxNumCompThreads https://www.mathworks.com/help/matlab/ref/maxnumcompthreads.html to set the desired max number of computational threads.
LASTN = maxNumCompThreads(N)
If this answer helps you, please remember to accept the answer.
  2 个评论
Walter Roberson
Walter Roberson 2023-12-14
? they are already doing that, but it isn't helping.
jessupj
jessupj 2023-12-14
maxNumCompThreads is ignored. the code snippet, when i run it, uses O(100) cores even despite the instruction maxNumCompThreads(k) for whatever value of k I throw at it

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Gaussian Process Regression 的更多信息

标签

产品


版本

R2023b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by