Are the returned timing values for the random and grid search implementations of sklearn in CPU or wall time? All I can find in the documentation is the following:
The mean_fit_time, std_fit_time, mean_score_time and std_score_time are all in seconds.
So I am assuming that would mean wall time as the default, is this correct?
After going through the source code...
Line 748:
self.cv_results_ = results
Okay, what is results?
Line 706:
results = self._format_results(
all_candidate_params, scorers, n_splits, all_out)
return results
Okay, what is all_out?
Line 703:
all_out.extend(out)
and then out is:
Line 682:
out = parallel(delayed(_fit_and_score)(clone(base_estimator),
and _fit_and_score is imported from sklearn.model_selection._validation.py so we look in there...
Line 543:
fit_time = time.time() - start_time
Aha! There we go. It's wall time, as Python's time.time()gives you the current Unix time.