[MNT] Update skbase to run sktime's contract validation tests

[felipeangelimvieira]

As discussed with @fkiraly, the previous version of skbase wasn't properly validating jax.numpy attributes (see sktime/skbase#324)

The tests are now being skipped, but can be properly executed with this update.

[fkiraly]

This specific issue should be resolved by sktime/skbase#323, right?

#324 is a "bonus" feature and we should keep the issue open, but for the purpose of prophetverse, this feels resolved?

[felipeangelimvieira]

Yes! It's working with #323. This week I expect to create a PR with some changes in code and without skipping the sktime contract tests. The tests passed in my local machine.

[felipeangelimvieira]

The test is creating some problems in the CI workflow... it seems to be a common issue with github actions (actions/runner-images#6680), and it may be related to CPU / memory usage. Have you ever experienced it in sktime, @fkiraly ?

Example of CI failure

[fkiraly]

Yes, and we have not fully diagnosed it.

Our suspicion is that this happens for memory intensive models, and the approach is selecting test parameters in get_test_params that lead to low memory and runtime requirements.

[fkiraly]

Plus, it seems that test_sktime_contract is being skipped, so the failure is coming from the other tests.

[felipeangelimvieira]

Plus, it seems that test_sktime_contract is being skipped, so the failure is coming from the other tests.

Locally, the tests are being executed partially but then skipped manually by a particular test case:

Output of pytest tests/sktime/test_sktime_check_estimator.py -vvrs

================================================================== short test summary info ===================================================================
SKIPPED [4] .venv/lib/python3.11/site-packages/sktime/forecasting/tests/test_all_forecasters.py:215: ForecastingHorizon with timedelta values is currently experimental and not supported everywhere
======================================================= 4 skipped, 1930 warnings in 619.14s (0:10:19) ========================================================

[fkiraly]

Should be addressed and explained by sktime/sktime#6587 - let's wait with closing this until the release though.