[ENH] Move Test Skip config for CNTCRegressor #8527
Conversation
| preds = self.model_.predict([X2, X, X], self.batch_size, **kwargs) | ||
| preds = np.squeeze(preds, axis=-1) | ||
| return preds | ||
| def _more_tags(self): |
There was a problem hiding this comment.
this is how it would work in sklearn, but not in sktime! here we use the _tags attribute (scroll up)
|
The only change in this PR was for CNTCRegressor’s test tags, but the failures are in ThetaForecaster, which wasn’t touched. Could this be a flaky or unrelated issue? |
|
I think this one is also suffering from memory leaks, and causing unrelated tests to crash. Could you also add the I will add the estimator to this issue: #8518 |
|
After adding "tests:vm": True, the CNTCRegressor tests trigger usage of SeqSelfAttention, but it seems it hasn't been registered with @register_keras_serializable() — causing failures during model loading. |
|
I am confused - why did this succeed in the normal CI? |
|
I'm also unsure why the normal CI passed, but the VM run fails after adding tests:vm. |
|
Hi, unrelated test (test_reconciler_fit_predict) is failing, even though my PR doesn’t touch that part of the code. |
|
Yes, the failures are unrelated! We can merge this. |
Reference Issues/PRs
[ENH] move test skip config from tests._config to estimator tags #8515
What does this implement/fix? Explain your changes.
It migrates test skip logic for the CNTCRegressor estimator from the centralized tests/_config.py file into the estimator itself using the _more_tags.
Does your contribution introduce a new dependency? If yes, which one?
No