MAINT activate common test sklearn - #647
Conversation
|
There is a failing test. It seems that the TableVectorizer no longer accepts 1D inputs. This is a good thing, but the changelog must be updated to reflect this, and the test updated. |
GaelVaroquaux
left a comment
GaelVaroquaux
left a comment
There was a problem hiding this comment.
One minor comment and then merge
| for j in range(X.shape[1]): | ||
| X[i, j] = str(X[i, j]) | ||
| elif _safe_tags(estimator, key="allow_nan"): | ||
| X = X.astype(np.float64) |
There was a problem hiding this comment.
The scikit-learn conversion will be np.float64 so it will be better for all common tests.
|
|
||
| def test_sklearn_compatible_GapEncoder(): | ||
| check_estimator(GapEncoder()) | ||
| if estimator.__class__.__name__ == "SkewedChi2Sampler": |
|
|
||
|
|
||
| def test_sklearn_compatible_SimilarityEncoder(): | ||
| check_estimator(SimilarityEncoder()) |
|
You have another test failing. The feature names have changed (I'm a bit surprised, I don't see where the change comes from, but it does not matter terribly). I don't mind the change of name. We should however update the test and maybe mention it in the changelog. |
|
The reason is that we now first |
|
However, this is a regression because this is now different from what the scikit-learn encoder are doing. |
Right, that might be a problem |
|
LGTM. Updating the branch with main and then merging. |
2 participants
Run scikit-learn common test.