[FIXED] How to limit sklearn GridSearchCV cpu usage?

Issue

I use GridSearchCV as follows:

gsearch_lgb = GridSearchCV(
    model(**self.model_params),
    param_grid=self.model_params,
    n_jobs=2,
    verbose=99,
    scoring=self.cv_scoring,
    cv=4,
)

But joblib still uses my all cores:

I also tried n_jobs=-19 since sklearn document said "For n_jobs below -1, (n_cpus + 1 + n_jobs) are used"

But still not working, all my cpus are used.

How should I modified my code to reduce CPU usage?

Solution

It turns out I have to set n_jobs as lgbm.Regressor parameter also, so the total CPU usage would be:

total_cpu = GridSeaechCV.fit n_jobs x lgbm.fit n_jobs

Answered By - Jim Chen