MNT: Update CI, add Python 3.11, fix sklearn issues #990
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Use latest versions.
PyTorch 2.0.1 should now run with Python 3.11. As to earlier PyTorch versions, it's not quite clear to me, let's see and if they don't, exclude them.
Makes errors from failing tests easier to read.
Something somewhere in sklearn changed that now requires the presence of the classes_ attribute on classifiers when scoring. That attribute is not always present. Now, more of an effort is made to infer the classes when a dataset is passed. Together with the previous commit that enables extraction of y from TensorDatasets, this makes a test pass that failed otherwise with the new sklearn version.
An sklearn error message that we checked was changed, resulting in the corresponding tests to fail. This commit fixes the error message checked in the test (which was already quite defensive beforehand).
|
Strange, according to this, PyTorch 2.0.1 should work with Python 3.11 but the CI fails with:
|
Otherwise, outputs are squished on GH action log.
@BenjaminBossan I don't believe the person that made that comment works on PyTorch. |
Yes, I had still hoped they were telling the truth :D |
Explicitly test that when fitting with a TensorDataset, the NeuralNetClassifier still manages to set the classes_ attribute.
ottonemo
approved these changes
Jul 6, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Updated description
checkoutandsetup-pythonGH actionsclasses_attribute is now required on classifiers for some stuff related to sklearn scoring, where previously it wasn't, which resulted in some errors when we use non-skorchDatasets (which so far didn't set the attribute). There is now a fix in place that extractsyfromTensorDatasets, making a failing test pass again. Other datasets could still fail, but I don't think we can write generic code that extractsyfrom all types of datasets (we want to avoid iterating over the dataset). For those, users would have to either use an older sklearn version or passyexplicitly tofit.Regarding the last point, I think the change of trying to infer the classes from a dataset should not make existing code break 🤞
*PyTorch 2.0.1 does not work, even though it supposedly should, see comment below