Looks like tests fail due to server issues, otherwise the change looks good to me.

Yes, trying to fix them together with @janvanrijn (this PR is part of it).

Looks like tests fail due to server issues, otherwise the change looks good to me.

I disagree.

The unit tests seem to assert that there are 2 public versions of the iris dataset on the test server, which is an incorrect assumption, especially after we reset the testserver.

The unit tests seem to assert that there are 2 public versions of the iris dataset on the test server

This PR changes this. The unit test now assumes that there are 2 public versions of the iris dataset on the live server.

That seems perfectly reasonable to me. This solves issue #858.

I meant the CI results on this PR came back negative, but those were due to server issues. This was unrelated to the change made for this PR, which was not a server issue.

can you point out specifically which failing unit tests you refer to?

Some might fail due to recent reset of test server.

Errors like this. I don't think they are persistent bugs, and might well have to do with the recent reset - I'm not saying the server is broken. But for the purpose of an openml-python PR I leave a comment indicating them as 'server issue' for later reference, indicating that we do see CI failures but don't expect them to be caused by the changes to openml-python.