As correctly pointed out by @Trybnetic we should make our defaults in n_jobs similar or the same to scikit-learn https://scikit-learn.org/stable/glossary.html#term-n_jobs
One option would be:
The default of n_jobs should be None. If n_jobs is None it should be set to min(n_local_cpus, 8) as this should in most cases give good speedups without losing the speed gain on 40 and more cpu machines. What do you think? I like the implementation and meaning of negative numbers as well, which is n_local_cpus - negative_number. n_jobs=-1 would be quite useful default on desktop machines.
Another options would be:
From my intuition, I would n_jobs expect either to be 1 or to use all cores (=-1) by default, with a tendency to use only one core if not specified. What do you think?
As correctly pointed out by @Trybnetic we should make our defaults in
n_jobssimilar or the same to scikit-learn https://scikit-learn.org/stable/glossary.html#term-n_jobsOne option would be:
The default of n_jobs should be None. If n_jobs is None it should be set to min(n_local_cpus, 8) as this should in most cases give good speedups without losing the speed gain on 40 and more cpu machines. What do you think? I like the implementation and meaning of negative numbers as well, which is n_local_cpus - negative_number. n_jobs=-1 would be quite useful default on desktop machines.
Another options would be:
From my intuition, I would n_jobs expect either to be 1 or to use all cores (=-1) by default, with a tendency to use only one core if not specified. What do you think?