[feature request] change verbosity level of _fit_and_score method

[AlJohri]

If I set verbose=2 within RandomizedSearchCV, I ideally want to see the steps that Parallel takes:

[Parallel(n_jobs=-1)]: Done   2 out of  10 | elapsed:   40.9s remaining:  2.7min

but I don't want to see the intermediate reporting of the CV search since that gets super noisy:

[CV] estimator__alpha=1.356430250055455e-06 ..........................
[CV] estimator__alpha=1.356430250055455e-06 ..........................
[CV] estimator__alpha=8.022557928125194e-06 ..........................
[CV] estimator__alpha=1.1751658491093436e-05 .........................
[CV] estimator__alpha=1.1751658491093436e-05 .........................
[CV] estimator__alpha=1.1751658491093436e-05 .........................
[CV] estimator__alpha=1.356430250055455e-06 ..........................
[CV] estimator__alpha=8.022557928125194e-06 ..........................

if we could change verbose > 1 to verbose > 2 this would solve my usecase.

scikit-learn/sklearn/model_selection/_validation.py

Lines 482 to 488 in acb8106

	if verbose > 1:
	if parameters is None:
	msg = ''
	else:
	msg = '%s' % (', '.join('%s=%s' % (k, v)
	for k, v in parameters.items()))
	print("[CV] %s %s" % (msg, (64 - len(msg)) * '.'))

I'd be happy to submit a PR if this makes sense.

Related:

• Use python logging to report on convergence progress it level info for long running tasks #78
• Why does _fit_and_score use print() instead of logging? #6487

[jnothman]

What does verbose=1 output? I would be okay with this kind of change, but we don't have any clear rules for it.

[AlJohri]

@jnothman verbose=1 outputs

[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.
[Parallel(n_jobs=-1)]: Done  30 out of  30 | elapsed: 46.5min finished

with nothing in the middle

[AlJohri]

@jnothman I filed a PR: #13091

[amueller]

closing as no reply in the PR. I don't think it's worth doing. Feel free to reply /reopen.