✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 04557fe. Link to the linter CI: here}

@scikit-learn-bot update lock-files --select-tag cuda

Something is weird no? Because the .lock file still contains pytorch from the pytorch channel :-/

Yeah I was hoping the @scikit-learn-bot magic comment would work but it looks like it does not, let me try to update locally and push.

Some files are downloaded directly from main and that includes update_environment_and_lock_files.py which you know make the @scikit-learn-bot not that useful ... see build log

This was done to prevent abuse, see #29505 (comment). This would need a closer look, in my opinion this makes the @scikit-learn-bot not super useful, you can only update the lock-files without changing anything but maybe it's safer to be conservative. The only use case I can see with the current behaviour is that you need the latest version of a package that is already in the lock-file and you can not wait until next Monday when lock-files get automatically updated. It can not be used in the vast majority of case, for example when you need a new package, or need to tweak an existing environment or lock-file.

By the way @betatim do you have automated alarms set-up for detecting CUDA activity in scikit-learn? Or you bumped into the PR by chance?

Looks like pytorch-gpu is installed and the test pass see build log

I am going to retrigger it with pytest -v just to check the gpu tests did run. Looks like they did run see build log you can see some test names with cuda with PASSED (and not SKIPPED):

conda/envs/sklearn/lib/python3.12/site-packages/sklearn/decomposition/tests/test_pca.py::test_pca_array_api_compliance[PCA(n_components=2,svd_solver='full',whiten=True)-check_array_api_input_and_values-torch-cuda-float64] PASSED [  1%]

By the way @betatim do you have automated alarms set-up for detecting CUDA activity in scikit-learn? Or you bumped into the PR by chance?

The 👁️ of CUDA makes my watch beep when there is a GPU issue or PR ;)

Just happened to be the top notification when I started looking at things

I am fine with keeping -v, since we are running only the array API tests the log isn't horribly longer.

More about the context: I wanted to at least try with -v to be reassured that somehow the CUDA tests were indeed running. I was slightly afraid that somehow the CUDA tests would be skipped without me noticing. Maybe part of it is that I am not very familiar with how array API tests are structured.