FIX allow copied parameters in __init__ for clone #22973
Conversation
|
Yes, I did predict this would happen, but decided not to mention it on the issue... I don't think it's a major concern, since we're only dealing here with badly behaved constructors, or else we could define a helper to suppress the ValueError. |
|
Or is your concern that the implementation might intentionally copy the array parameter, or realise a parameter stored in a C struct as an array? If so, let's use a helper. |
|
I have a similar helper here scikit-learn/sklearn/model_selection/_search.py Lines 50 to 73 in 3d9fd3f |
|
Slightly confused by this bit: if v1 is not v2:
return False
with warnings.catch_warnings():
warnings.simplefilter("ignore")
try:
if v1 != v2:
return False
except ValueError:
return FalseWhat should the input be for the code to reach if v1 is v2:
continue
with warnings.catch_warnings():
warnings.simplefilter("ignore")
try:
if v1 != v2:
return False
except ValueError:
return False |
|
I agree that doesn't look right... Makes me wonder about my test coverage. |
| pass | ||
|
|
||
| # Register global_random_seed plugin if it is not already registered | ||
| if not config.pluginmanager.hasplugin("sklearn.tests.random_seed"): |
There was a problem hiding this comment.
Is this a merge error? This change is already on main.
| def __init__(self, a=np.array([0])): | ||
| self.a = a.copy() |
|
Is this PR still relevant since we now have #24568 ? |
Fixes #22857
There is an issue I hadn't realized: comparing ndarrays raises a
ValueErrorand not aRuntimeError. So if the estimator does a copy of an ndarray input argument, clone would fail with a different error. I'm not sure how to proceed.Also pinging @amueller (you wrote the test which fails with this test I think)