[ENH] python 3.11 support

[fkiraly]

update, Feb 23, 2023: numba is now fully isolated, and sktime release 0.16.1 has been released with experimental 3.11 support.
Issues that remain to fix before full support in 0.17.0:

• soft dependencies need to be checked with 3.11, and restrictions removed where appropriate [MNT] add soft dependencies on python 3.11 that are 3.11 compatible #4269
• conda issues with 3.11 need to be fixed python 3.11 support conda-forge/sktime-feedstock#66
  this is waiting for migration, and is not necessarily blocking 0.17.0

---

Umbrella issue for discussing python 3.11 support, to direct frequent questions to.

Current blockers to the "naive" solution of releasing sktime for 3.11 as-is, as far as I know:

• numba is a core dependency and not python 3.11 compatible
• a number of soft dependencies do not have python 3.11 support

Potential solutions:

• soft dependencies are all isolated, so conditional install in pyproject.toml would fix the issue. This would require us to check python 3.11 support for soft dependencies periodically and potentially remove the condition at a later point in time.
• numba could be solved by "wait for python 3.11 support". When that happens is unclear.
• or, we can turn numba into a soft dependency right now, and deal with it as with the other soft dependencies. This is entirely feasible - the PR here could be merged and enable python 3.11 EXPERIMENTAL: [MNT] python 3.11 compatibility with numba as soft dependency #3856. But, it is a very contentious issue with some core developers.
• a third option keeps numba as core dependency, but makes it optional only on python 3.11, until it becomes compatible. This option is implemented here, ready for merge and enable: [MNT] python 3.11 compatibility, with numba as core dependency on 3.7-3.10 #4000. As it turns out, this is also very contentious.

[fkiraly]

FYI @alexanderreschjedox, I tried to summarize the situation above.

[jplatipus]

The github page should be updated: currently https://github.com/sktime/sktime says sktime works with python 3.11, but I get a conflict report:

(sktimeTests) C:\Users\jacqu>conda install -c conda-forge sktime
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.
Collecting package metadata (repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: |
Found conflicts! Looking for incompatible packages.
This can take several minutes. Press CTRL-C to abort.
failed

UnsatisfiableError: The following specifications were found
to be incompatible with the existing python installation in your environment:

Specifications:

• sktime -> python[version='>=3.6,<3.7.0a0|>=3.7,<3.8.0a0|>=3.8,<3.9.0a0|>=3.9,<3.10.0a0']

Your python: python=3.11

If python is on the left-most side of the chain, that's the version you've asked for.
When python appears to the right, that indicates that the thing on the left is somehow
not available for the python version you are constrained to. Note that conda will not
change your python version to a different minor version unless you explicitly specify
that.

[fkiraly]

@jplatipus, as far as I know sktime works on 3.11 from version 0.16.1 onwards. Support is currently experimental.

3.11 compatibility is tested in the CI/CD pipeline, and I just checked pip install from conda and I do not get the message.
pip (e.g., from conda) is how the majority of users are installing sktime, and this works as intended.

Now you are using conda-forge, i.e., conda install -c conda-forge sktime.
I just tried in a python 3.11 environment, on Windows OS, and it seems to get stuck (not getting an UnsatisfiableError)

I will open an issue to track this.

[fkiraly]

Issue is here: #4254