hebo take up too much cpus

[Shaofei-Li]

The main process of HEBO take up 1600% cpus, which makes other parts of the program run very slowly. Is there any way to reduce the cpu usage of hebo? Thanks!

[Alaya-in-Matrix]

I think it's because numpy and pytorch used multi-threading for their computation, CPU usage should be limited if you disable numpy/torch multi-threading.

https://pytorch.org/docs/stable/generated/torch.set_num_threads.html

https://stackoverflow.com/questions/30791550/limit-number-of-threads-in-numpy

[Alaya-in-Matrix]

I just wrote a demo, adding the import os; os.environ..... statments reduces CPU usage from 800% to 100% on my own machine.

import os
os.environ['OMP_NUM_THREADS'] = '1'
os.environ['MKL_NUM_THREADS'] = '1'

from tqdm import trange
from hebo.optimizers.hebo import HEBO
from hebo.benchmarks.synthetic_benchmarks import RosenbrockDummy

if __name__ == '__main__':
    prob = RosenbrockDummy(10)
    opt  = HEBO(prob.space)
    tbar = trange(100)
    for i in tbar:
        rec = opt.suggest()
        y   = prob(rec)
        opt.observe(rec, y)
        tbar.set_description(f'Iter {i}, y = {y.min():.2f}, best_y = {opt.y.min():.2f}')

[Alaya-in-Matrix]

To me, numpy multi-threading does not seem to offer observable speed-up, so I actually just set the two environment variables to 1 in my .bashrc