[RFC] Drop Python 3.6 support

We will gracefully stop the support for Python 3.6 after the EOL (23 Dec 2021). https://devguide.python.org/#branchstatus

For now, we are not going to force users to upgrade Python (i.e. we currently don’t have a plan to update python_requires to be >=3.7 immediately after EOL). However, third-party libraries drop/will drop Python 3.6 before/after EOL and these may cause problems when using Optuna. We’ll gradually stop dealing with such problems and will recommend users upgrade their Python.

[not522]

We decide to stop Python 3.6 support for integrations and optuna-examples but keep supporting it in pure Optuna uses. It means that we will remove Python 3.6 tests from the CI of the integrations and the examples but remain in the other CIs.

The maintenance cost of testing Python 3.6 with third-party libraries is increasing because some of them have stopped supporting Python 3.6. However, some Optuna users still use Python 3.6, and then we do not update python_requires to be >=3.7 now.

[nzw0301]

Note that the next scipy drops python 3.7. Python 3.6 has already been dropped. I think numpy is the same.

• ref: https://github.com/scipy/scipy/blob/master/setup.py#L30-L31.
• https://numpy.org/neps/nep-0029-deprecation_policy.html

---

Third-party libraries

• allennlp 2.8.0 drops python 3.6 support
• botorch does not support 3.6 for while.
• scikit-learn drops 3.6 support since v1.0.0
• LightGBM drops python 3.6 support https://github.com/microsoft/LightGBM/pull/4891
• PyTorch 1.11.0 drops Python 3.6