Restricting core usage

[PaMartini]

Is there a way to specify the cores that are used during training? I am using the Python interface to Somoclu. I tried:

import psutil
import os

cores = [0,1,3]

p = psutil.Process(os.getpid())
p.cpu_affinity(cores)

os.environ["OMP_PROC_BIND"] = 'TRUE'
os.environ["OMP_PLACES"] = ','.join(f'{{{core}}}' for core in cores)

... script that uses Somoclu to train a SOM ...

But this does not work for me ... (I guess the parallelization of Somoclu ignores the set parameters.) (Also, I know that this is possible via taskset, but I would prefer to specify the cores in the python code.)

[xgdgsc]

What if setting a global env variable instead of inside python? Or start a new process with those env variables?

[PaMartini]

Isn't

os.environ["OMP_PROC_BIND"] = 'TRUE'
os.environ["OMP_PLACES"] = ','.join(f'{{{core}}}' for core in cores)

setting a global env variable?

How would you do this: "Or start a new process with those env variables?"

The reason for my request was that i would like to have the cores on which SOM training with Somoclu is run as an input parameter to my python function. If this is not possible due to Somoclus parallelization beeing out of reach for Pyton, I guess the best option would be to use taskset from the commandline.

[xgdgsc]

Like https://stackoverflow.com/a/4453495/1136027