✔️ Linting Passed

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

_{Generated for commit: e9fd011. Link to the linter CI: here}

looks like the failure comes from openml being down for some time.

The failure is due to openml being down. I'll retry later when it's up.

I suppose that we do not experience the problem on the other CI builds because of the CI cache key that is derived from the lock files?

If so that is a very nice side effect of using lock files :) /cc @lesteve.

I don't think lock-files are involved, it is more that most builds don't actually download data from OpenML 😉. Even CircleCI builds won't download data from OpenML unless you use [doc build] or modify an example. Cache may also help in some cases.

FYI, there is a bit of info of the OpenML status on their Slack. Last update was yesterday (January 15th):

Update: "TU/e expects to have most of its network and network-bound systems secure and available again before the weekend." (source)
We do not know if OpenML is included in "most", let's hope so

We now also get HTTP 403 errors when calling fetch_california_housing in 2 doctests (in compose.rst and getting_started.rst). Maybe we could rewrite those doctests to avoid relying on the internet.

This problem is probably caused by an anti-DOS mechanism of the figshare hosting platforms with a denylist on a range of IP addresses from cloud hosted machines sometime used by our CI workers. We experienced a similar problem in the past:

• Getting HTTPError: HTTP Error 403: Forbidden when trying to load California Housing dataset #28297

We now also get HTTP 403 errors when calling fetch_california_housing in 2 doctests (in compose.rst and getting_started.rst). Maybe we could rewrite those doctests to avoid relying on the internet.

I had seen the doctest failures and just thought "oh well another failure due to OpenML" but indeed it's not fetched from OpenML 😅 ...

Whatever is easier:

• either we skipped those doctests when SKLEARN_SKIP_NETWORK_TESTS is unset. I think we are only running network test in a scheduled nightly right now.
• rewrite the doctests without relying on the internet
• something even simpler if it exists

Maybe I am over-cautious but I would not merge this without making sure that the full doc build works, which we can not do for now since OpenML is still not fully functional. Maybe I am missing something or maybe you ran the doc build locally?

I have disabled auto-merge for now.

OK I now had a look at the diff and it seems like it's very minimal for the doc builds so I guess YOLO while OpenML is still problematic and we still can 🤠